gcc/ChangeLog

   1 2018-04-30  Jason Merrill  <jason@redhat.com>
   2
   3         * tree.c (build_clobber): New.
   4         * tree.h: Declare it.
   5         * gimplify.c (gimplify_bind_expr, gimplify_target_expr): Use it.
   6
   7 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
   8
   9         * diagnostic-show-locus.c (layout::layout): Update for
  10         location_get_source_line returning a char_span.
  11         (struct char_span): Move to input.h.
  12         (struct correction): Update for fields in char_span becoming
  13         private.
  14         (struct source_line): Update for location_get_source_line
  15         returning a char_span.
  16         (layout::print_line): Likewise.
  17         * edit-context.c (edited_file::print_content): Likewise.
  18         (edited_file::print_diff_hunk): Likewise.
  19         (edited_file::print_run_of_changed_lines): Likewise.
  20         (edited_file::get_num_lines): Likewise.
  21         (edited_line::edited_line): Likewise.
  22         * final.c (asm_show_source): Likewise.
  23         * input.c (location_get_source_line): Convert return type
  24         from const char * to char_span, losing the final "line_len"
  25         param.
  26         (dump_location_info): Update for the above.
  27         (get_substring_ranges_for_loc): Likewise.  Use a char_span
  28         when handling the literal within the line.
  29         (test_reading_source_line): Update for location_get_source_line
  30         returning a char_span.
  31         * input.h (class char_span): Move here from
  32         diagnostic-show-locus.c, converting from a struct to a class.
  33         Make data members private.
  34         (char_span::operator bool): New.
  35         (char_span::length): New.
  36         (char_span::get_buffer): New.
  37         (char_span::operator[]): New.
  38         (char_span::subspan): Make const.
  39         (char_span::xstrdup): New.
  40         (location_get_source_line): Convert return type from const char *
  41         to char_span, losing the final "line_size" param.
  42
  43 2018-04-30  Jan Hubicka  <jh@suse.cz>
  44
  45         * lto-wrapper.c (ltrans_priorities): New static var.
  46         (cmp_priority): New.
  47         (run_gcc): Read priorities and if doing parallel build order
  48         the Makefile by them.
  49
  50 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  51
  52         * input.h (builtins_location_check): Convert to a STATIC_ASSERT.
  53
  54 2018-04-30  Richard Biener  <rguenther@suse.de>
  55
  56         * tree-cfg.c (verify_address): Remove base argument, add
  57         flag whether to check TREE_ADDRESSABLE and do that.
  58         (verify_expr): Remove.
  59         (verify_types_in_gimple_reference): Add pieces from verify_expr.
  60         (verify_gimple_assign_single): Likewise.
  61         (verify_gimple_switch): Likewise.
  62         (verify_expr_location_1): Dereference tp once.  Add (disabled)
  63         piece from verify_expr.
  64         (verify_gimple_in_cfg): Do not call verify_expr on all ops.
  65
  66 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
  67
  68         * config/arc/linux.h (CLEAR_INSN_CACHE): Define.
  69
  70 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
  71
  72         * config/arc/arc-protos.h (prepare_extend_operands): Remove.
  73         (small_data_pattern): Likewise.
  74         (arc_rewrite_small_data): Likewise.
  75         * config/arc/arc.c (LEGITIMATE_SMALL_DATA_OFFSET_P): Remove.
  76         (LEGITIMATE_SMALL_DATA_ADDRESS_P): Likewise.
  77         (get_symbol_alignment): New function.
  78         (legitimate_small_data_address_p): Likewise.
  79         (legitimate_scaled_address): Update, call
  80         legitimate_small_data_address_p.
  81         (output_sdata): New static variable.
  82         (arc_print_operand): Update how we handle small data operands.
  83         (arc_print_operand_address): Likewise.
  84         (arc_legitimate_address_p): Update, use
  85         legitimate_small_data_address_p.
  86         (arc_rewrite_small_data_p): Remove.
  87         (arc_rewrite_small_data_1): Likewise.
  88         (arc_rewrite_small_data): Likewise.
  89         (small_data_pattern): Likewise.
  90         (compact_sda_memory_operand): Update to use
  91         legitimate_small_data_address_p and get_symbol_alignment.
  92         (prepare_move_operands): Don't rewite sdata pattern.
  93         (prepare_extend_operands): Remove.
  94         * config/arc/arc.md (zero_extendqihi2): Don't rewrite sdata
  95         pattern.
  96         (zero_extendqisi2): Likewise.
  97         (zero_extendhisi2): Likewise.
  98         (extendqihi2): Likewise.
  99         (extendqisi2): Likewise.
 100         (extendhisi2): Likewise.
 101         (addsi3): Likewise.
 102         (subsi3): Likewise.
 103         (andsi3): Likewise.
 104         * config/arc/constraints.md (Usd): Change it to memory constraint.
 105
 106 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
 107
 108         * config/arc/arc.c (arc_split_move): Allow signed 6-bit constants
 109         as source of std instructions.
 110         * config/arc/arc.md (movsi_insn): Update pattern predicate to
 111         allow 6-bit constants as source for store instructions.
 112         (movdi_insn): Update instruction pattern to allow 6-bit constants
 113         as source for store instructions.
 114
 115 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
 116
 117         * doc/invoke.texi (-fdebug-types-section): Fix grammar.
 118
 119 2018-04-30  Nathan Sidwell  <nathan@acm.org>
 120             Sandra Loosemore <sandra@codesourcery.com>
 121
 122         * dumpfile.c (dump_open): Allow '-' for stdout.
 123         * doc/invoke.texi (Developer Options): Document dump filename
 124         determination early.  Document stdin/stdout selection.
 125
 126 2018-04-30 Andrew Sadek  <andrew.sadek.se@gmail.com>
 127
 128         Microblaze Target: PIC data text relative
 129
 130         * config/microblaze/microblaze.opt: add new option -mpic-data-text-rel.
 131         * config/microblaze/microblaze-protos.h (microblaze_constant_address_p):
 132         Add declaration.
 133         * gcc/config/microblaze/microblaze.h (microblaze_constant_address_p):
 134         CONSTANT_ADDRESS_P definition to microblaze_constant_address_p.
 135         * config/microblaze/microblaze.c (TARGET_PIC_DATA_TEXT_REL):
 136         New addressing mode for data-text relative position indepenedent code.
 137         (microblaze_classify_unspec): add 'UNSPEC_TEXT' case ->
 138         'ADDRESS_SYMBOLIC_TXT_REL'.
 139         (microblaze_classify_address): Add handling for UNSPEC + CONST_INT.
 140         (microblaze_legitimate_pic_operand): Exclude function calls from
 141         pic operands in case of TARGET_PIC_DATA_TEXT_REL option.
 142         (microblaze_legitimize_address): Generate 'UNSPEC_TEXT' for all possible
 143         addresses cases.
 144         (microblaze_address_insns): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
 145         (print_operand): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
 146         (print_operand_address): Add 'ADDRESS_SYMBOLIC_TXT_REL' case + handling
 147         for 'address + offset'.
 148         (microblaze_expand_prologue): Add new function prologue call for
 149         'r20' assignation.
 150         (microblaze_asm_generate_pic_addr_dif_vec): Override new target hook
 151         'TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC' to disable address diff vector
 152         table in case of TARGET_PIC_DATA_TEXT_REL.
 153         (expand_pic_symbol_ref): Add handling for 'UNSPEC_TEXT'.
 154         * gcc/config/microblaze/microblaze.md (TARGET_PIC_DATA_TEXT_REL):
 155         Add new macros 'UNSPEC_TEXT',
 156         'UNSPEC_SET_TEXT' + add rule for setting r20 in function prologue
 157         + exclude function calls from 'UNSPEC_PLT' in case of data text
 158         relative mode.
 159         * doc/tm.texi.in (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
 160         new target hook for generating address diff vector tables in case of
 161         flag_pic.
 162         * doc/tm.texi : Regenerate.
 163         * stmt.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Append new condition
 164         'targetm.asm_out.generate_pic_addr_diff_vec' to flag_pic in case
 165         of addr diff vector generation.
 166         * target.def (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
 167         target hook definition.
 168         * targhooks.h, gcc/targhooks.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC):
 169         Add default function for generate_pic_addr_diff_vec -> flag_pic.
 170         * doc/invoke.texi (Add new pic option): Add new microblaze pic
 171         option for data text relative.
 172
 173 2018-04-30  Richard Biener  <rguenther@suse.de>
 174
 175         * tree-chrec.h (evolution_function_is_constant_p): Remove
 176         redundant check.
 177         * tree-cfg.c (tree_node_can_be_shared): Re-order checks.
 178
 179 2018-04-30  Richard Biener  <rguenther@suse.de>
 180
 181         PR bootstrap/85571
 182         * dwarf2out.c (gen_producer_string): Ignore -fchecking[=].
 183
 184 2018-04-30  Richard Biener  <rguenther@suse.de>
 185
 186         PR tree-optimization/28364
 187         PR tree-optimization/85275
 188         * tree-ssa-loop-ch.c (ch_base::copy_headers): Stop after
 189         copying first exit test.
 190
 191 2018-04-28  Mark Wielaard  <mark@klomp.org>
 192
 193         * dwarf2out.c (dwarf2out_finish): Add .debug_addr table header for
 194         dwarf_version >= 5.
 195         (dwarf_AT): Handle DW_AT_addr_base.
 196         (add_top_level_skeleton_die_attrs): Use dwarf_AT for DW_AT_addr_base.
 197
 198 2018-04-28  Uros Bizjak  <ubizjak@gmail.com>
 199
 200         PR target/84431
 201         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask): New pattern.
 202         (*ashl<dwi>3_doubleword_mask_1): Ditto.
 203         (*<shift_insn><dwi>3_doubleword_mask): Ditto.
 204         (*<shift_insn><dwi>3_doubleword_mask_1): Ditto.
 205
 206 2018-04-28  Richard Biener  <rguenther@suse.de>
 207
 208         * tree-cfg.c (verify_gimple_phi): Take a gphi * argument.
 209         (verify_gimple_in_cfg): Rename visited_stmts to visited_throwing_stmts
 210         to reflect use.  Only add interesting stmts.
 211
 212 2018-04-27  Martin Jambor  <mjambor@suse.cz>
 213
 214         PR ipa/85549
 215         * ipa-cp.c (find_aggregate_values_for_callers_subset): Make sure
 216         the jump function allows for passing through aggregate values.
 217
 218 2018-04-27  David Malcolm  <dmalcolm@redhat.com>
 219
 220         * input.h (in_system_header_at): Convert from macro to inline
 221         function.
 222         (from_macro_expansion_at): Likewise.
 223         (from_macro_definition_at): Likewise.
 224
 225 2018-04-27  Jeff Law  <law@redhat.com>
 226
 227         * config.gcc: Mark tile* targets as deprecated/obsolete.
 228
 229 2018-04-27  Richard Biener  <rguenther@suse.de>
 230
 231         * config/aarch64/aarch64.c: Simplify ap.__stack advance and
 232         fix for ILP32.
 233
 234 2018-04-27  Richard Biener  <rguenther@suse.de>
 235
 236         * tree-cfg.c (verify_expr): Make dead code hit gcc_unreachable.
 237
 238 2018-04-27  Uros Bizjak  <ubizjak@gmail.com>
 239
 240         * config/i386/i386.md (*movti_internal): Substitute Ye constraint
 241         with Yd constraint. Set "preferred_for_speed" attribute from
 242         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for alternatives
 243         with Yd constraint.
 244         (*movdi_internal): Ditto.
 245         (movti_interunit splitters): Remove
 246         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC from insn condition.
 247         (movdi_interunit splitters): Ditto.
 248         * config/i386/constraints.md (Ye): Remove.
 249         (Yd): Do not depend on TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC.
 250
 251 2018-04-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 252
 253         PR target/85512
 254         * config/aarch64/constraints.md (Usg): Limit to 31.
 255         (Usj): Limit to 63.
 256
 257 2018-04-27  Jakub Jelinek  <jakub@redhat.com>
 258
 259         PR tree-optimization/85529
 260         * tree-ssa-reassoc.c (optimize_range_tests_var_bound): Add FIRST_BB
 261         argument.  Don't call get_nonzero_bits if opcode is ERROR_MARK_NODE,
 262         rhs2 def stmt's bb is dominated by first_bb and it isn't an obvious
 263         zero extension or masking of the MSB bit.
 264         (optimize_range_tests): Add FIRST_BB argument, pass it through
 265         to optimize_range_tests_var_bound.
 266         (maybe_optimize_range_tests, reassociate_bb): Adjust
 267         optimize_range_tests callers.
 268
 269 2018-04-26  Richard Biener  <rguenther@suse.de>
 270             Jakub Jelinek  <jakub@redhat.com>
 271
 272         * cgraph.h (symbol_table): Just declare debug method here.
 273         * symtab.c (symbol_table::debug): Define.
 274
 275 2018-04-26  Eric Botcazou  <ebotcazou@adacore.com>
 276
 277         * loop-invariant.c (may_assign_reg_p): Return false for frame pointer.
 278
 279 2018-04-26  Uros Bizjak  <ubizjak@gmail.com>
 280
 281         * config/i386/i386.md ("isa" attribute): Add x64_sse2.
 282         ("enabled" attribute): Handle x64_sse2 "isa" attribute.
 283         (*movdi_internal): Substitute Yi and Yj constraint with x
 284         and Ym and Yn constraint with y constraint.  Update "isa"
 285         attribute and set "preferred_for_speed" attribute from
 286         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for updated alternatives.
 287         (*movsi_internal): Ditto.
 288         (*movdf_internal): Ditto.
 289         (*movsf_internal): Ditto.
 290         (*zero_extendsidi2): Ditto.
 291         * config/i386/sse.md (vec_set<mode>_0): Ditto.
 292         (sse2_loadld): Ditto.
 293         (*vec_extract<ssevecmodelower>_0): Ditto.
 294         (*vec_extractv4si_0_zext_sse4): Ditto.
 295         (vec_concatv2di): Ditto.
 296         (*vec_dup<mode>): Ditto.
 297         * config/i386/mmx.md (*mov<mode>_internal): Ditto.
 298         * config/i386/constraints.md (Yi): Remove.
 299         (Yj): Remove.
 300         (Ym): Remove.
 301         (Yn): Remove.
 302
 303 2018-04-26  Nathan Sidwell  <nathan@acm.org>
 304
 305         * dumpfile.c (dump_open): New.
 306         (dump_open_alternate_stream, dump_start, dump_begin): Call it.
 307         (dump_finish): Detect stdio/stderr by value not name.
 308
 309 2018-04-26  Jonathan Wakely  <jwakely@redhat.com>
 310
 311         * doc/invoke.texi (-Wreturn-type): Document default status for C++.
 312
 313 2018-04-26  Tom de Vries  <tom@codesourcery.com>
 314
 315         PR target/84952
 316         * config/nvptx/nvptx.c (verify_neutering_jumps)
 317         (verify_neutering_labels): New function
 318         (nvptx_single): Use verify_neutering_jumps and verify_neutering_labels.
 319
 320 2018-04-26  Tom de Vries  <tom@codesourcery.com>
 321
 322         PR target/84025
 323         * config/nvptx/nvptx.c (needs_neutering_p): New function.
 324         (nvptx_single): Use needs_neutering_p to skip over insns that do not
 325         need neutering.
 326
 327 2018-04-26  Richard Biener <rguenther@suse.de>
 328             Tom de Vries  <tom@codesourcery.com>
 329
 330         PR lto/85422
 331         * lto-streamer-out.c (output_function): Fixup loops if required to match
 332         discovery done in the reader.
 333
 334 2018-04-26  Richard Biener  <rguenther@suse.de>
 335
 336         PR tree-optimization/85116
 337         * tree-ssa-loop-ch.c (do_while_loop_p): A do-while loop should
 338         have a loop exit from the single latch predecessor.  Remove
 339         case of header with just condition.
 340         (ch_base::copy_headers): Exclude infinite loops from any
 341         processing.
 342         (pass_ch::execute): Record exits.
 343
 344 2018-04-26  Richard Biener  <rguenther@suse.de>
 345
 346         * tree-vect-data-refs.c (vect_get_data_access_cost): Get
 347         prologue cost vector and pass it to vect_get_load_cost.
 348         (vect_get_peeling_costs_all_drs): Likewise.
 349         (vect_peeling_hash_get_lowest_cost): Likewise.
 350         (vect_enhance_data_refs_alignment): Likewise.
 351
 352 2018-04-26  Richard Biener  <rguenther@suse.de>
 353
 354         PR middle-end/85450
 355         * tree-cfg.c (verify_gimple_assign_unary): Restore proper
 356         checking of integer<->pointer conversions.
 357         * omp-expand.c (expand_omp_for_static_nochunk): Avoid
 358         sign-/zero-extending pointer types.
 359         (expand_omp_for_static_chunk): Likewise.
 360
 361 2018-03-22  Hans-Peter Nilsson  <hp@axis.com>
 362             Jean Lee  <xiaoyur347@gmail.com>
 363
 364         * config/mips/mips.c (mips_asan_shadow_offset): New function.
 365         (TARGET_ASAN_SHADOW_OFFSET): Define.
 366         * config/mips/mips.h (FRAME_GROWS_DOWNWARD): Augment to also be
 367         true for -fsanitize=address.
 368
 369 2018-04-25  Mark Wielaard  <mark@klomp.org>
 370
 371         * dwarf2out.c (file_info_cmp): Sort longer dir prefixes before
 372         shorter ones.
 373
 374 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
 375
 376         * config/i386/i386.md (*x86_mov<mode>cc_0_m1): Use type "alu1" rather
 377         than "alu", remove explicit "memory" and "imm_disp" attributes.
 378         (*x86_mov<mode>cc_0_m1_se, *x86_mov<mode>cc_0_m1_neg): Likewise.
 379
 380         PR middle-end/85414
 381         * simplify-rtx.c (simplify_unary_operation_1) <case SIGN_EXTEND,
 382         case ZERO_EXTEND>: Pass SUBREG_REG (op) rather than op to
 383         gen_lowpart_no_emit.
 384
 385 2018-04-25  Sebastian Peryt  <sebastian.peryt@intel.com>
 386
 387         PR target/85473
 388         * config/i386/i386.c (ix86_expand_builtin): Change memory
 389         operand to XI, extend p0 to Pmode.
 390         * config/i386/i386.md: Change unspec volatile and operand
 391         1 mode to XI, change operand 0 mode to P.
 392
 393 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 394
 395         * config/nds32/nds32-predicates.c (nds32_can_use_bclr_p): Mask with
 396         GET_MODE_MASK before any checking.
 397         (nds32_can_use_bset_p): Likewise.
 398         (nds32_can_use_btgl_p): Likewise.
 399
 400 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 401
 402         * config/nds32/nds32-doubleword.md: New define_split pattern for
 403         illegal register number.
 404
 405 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 406
 407         * config/nds32/nds32.c (nds32_print_operand): Set op_value ealier.
 408
 409 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 410
 411         * config/nds32/nds32.h (ASM_APP_ON): Add missing newline character.
 412
 413 2018-04-25  Richard Biener  <rguenther@suse.de>
 414
 415         * lto-streamer.h (LTO_major_version): Bump to 8.
 416
 417 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
 418
 419         * BASE-VER: Set to 9.0.0.
 420
 421 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
 422
 423         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
 424         in __abskf2 and __powikf2.
 425
 426 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 427
 428         PR target/85512
 429         * config/aarch64/constraints.md (Usg, Usj): New constraints.
 430         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
 431         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
 432         Use the above on operand 2.  Reindent.
 433         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
 434
 435 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
 436
 437         PR target/85485
 438         * common/config/i386/i386-common.c (ix86_handle_option): Don't
 439         handle OPT_mcet.
 440         * config/i386/i386.opt (mcet): Removed.
 441         * doc/install.texi: Remove -mcet documentation.
 442         * doc/invoke.texi: Likewise.
 443
 444 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
 445
 446         PR target/85485
 447         * doc/install.texi: Remove -mcet from bootstrap-cet.
 448
 449 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 450
 451         PR target/85511
 452         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
 453         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
 454         if TARGET_64BIT.
 455
 456         PR target/85503
 457         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
 458         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
 459         containing a CONST_VECTOR.
 460
 461 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
 462
 463         * doc/install.texi: Update newlib dependency for nvptx.
 464
 465 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 466
 467         PR target/85508
 468         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
 469         instead of INTVAL when shifting x left.
 470
 471 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
 472
 473         PR tree-optimization/85478
 474         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
 475         vect_grouped_store_supported for single element vectors.
 476
 477 2018-04-24  Richard Biener  <rguenther@suse.de>
 478
 479         PR target/85491
 480         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
 481         load cost increase to the case of non-constant step.
 482
 483 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 484
 485         PR target/84828
 486         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
 487         destination if any_malformed_asm.
 488
 489 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
 490
 491         PR middle-end/85496
 492         * expr.c (store_field): In the bitfield case, if the value comes from
 493         a function call and is returned in registers by means of a PARALLEL,
 494         do not change the mode of the temporary unless BLKmode and VOIDmode.
 495
 496 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
 497
 498         PR rtl-optimization/85423
 499         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
 500         dependencies to debug insns when the previous insn is non-debug.
 501
 502 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
 503
 504         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
 505         enums into a single definition.
 506         (fls): Fix predicates and printing.
 507         (seti): Likewise.
 508
 509 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
 510
 511         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
 512         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
 513         and short u6 immediate.
 514         (check_if_valid_sleep_operand): Remove.
 515         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
 516
 517 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 518
 519         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
 520         flag_always_save_lp condition.
 521         * config/nds32/nds32.opt (malways-save-lp): New option.
 522
 523 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 524
 525         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
 526         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
 527         * config/nds32/nds32.h
 528         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
 529         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
 530
 531 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 532
 533         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
 534         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
 535
 536 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 537             Chung-Ju Wu  <jasonwucj@gmail.com>
 538
 539         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
 540         Declare.
 541         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
 542         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
 543
 544 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 545
 546         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
 547
 548 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 549
 550         * config/nds32/nds32-protos.h (nds32_data_alignment,
 551         nds32_local_alignment): Declare.
 552         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
 553         nds32_local_alignment): New functions.
 554         (TARGET_CONSTANT_ALIGNMENT): Define.
 555         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
 556
 557 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 558
 559         * config/nds32/nds32.c
 560         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
 561         (TARGET_MODES_TIEABLE_P): Likewise.
 562
 563 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 564
 565         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
 566         level Ofast and Og.
 567
 568 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
 569             Chung-Ju Wu  <jasonwucj@gmail.com>
 570
 571         * config/nds32/constants.md (unspec_volatile_element): Add enum values
 572         for unaligned access.
 573         * config/nds32/nds32-intrinsic.c: Implementation of expanding
 574         unaligned access.
 575         * config/nds32/nds32-intrinsic.md: Likewise.
 576         * config/nds32/nds32_intrinsic.h: Likewise.
 577         * config/nds32/nds32.h (nds32_builtins): Likewise.
 578         * config/nds32/nds32.opt (munaligned-access): New option.
 579         * config/nds32/nds32.c (nds32_asm_file_start): Display
 580         flag_unaligned_access status.
 581
 582 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 583
 584         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
 585         -mno-relax is present.
 586         * config/riscv/linux.h (LINK_SPEC): Ditto.
 587
 588 2018-04-20  Martin Sebor  <msebor@redhat.com>
 589
 590         PR c/85365
 591         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
 592         for null pointers.
 593         (gimple_fold_builtin_stxcpy_chk): Same.
 594         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
 595
 596 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
 597
 598         PR target/85456
 599         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
 600         __powikf2 when long double is IEEE 128-bit.
 601
 602 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 603
 604         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
 605         step to make sure stack always aligned.
 606
 607 2018-04-20  Carl Love  <cel@us.ibm.com>
 608
 609         PR target/83402
 610         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
 611         size check for arg0.
 612
 613 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
 614             Tom de Vries  <tom@codesourcery.com>
 615
 616         PR target/85445
 617         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
 618         Emit insns for calls too.
 619         (nvptx_find_par): Always look for worker-level predecessor insn.
 620         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
 621         calls.
 622         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
 623         (nvptx_process_pars): Propagate frames for calls.
 624
 625 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
 626
 627         PR target/85469
 628         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
 629         Removed.
 630         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
 631         (ix86_handle_option): Don't handle OPT_mibt.
 632         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
 633         __SHSTK__.
 634         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
 635         has_ibt and ibt.
 636         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
 637         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
 638         (ix86_target_macros): Define __CET__ with flag_cf_protection
 639         for -fcf-protection.
 640         * config/i386/i386.c (isa2_opts): Remove -mibt.
 641         * config/i386/i386.h (TARGET_IBT): Removed.
 642         (TARGET_IBT_P): Likewise.
 643         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
 644         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
 645         * config/i386/i386.opt (mcet): Update help message.
 646         (mshstk): Likewise.
 647         (mibt): Removed.
 648         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
 649         -mcet as an alias for -mshstk.
 650
 651 2018-04-20  Richard Biener <rguenther@suse.de>
 652
 653         PR middle-end/85475
 654         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
 655         complexity by forcing a single use of the multiply operand.
 656
 657 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 658
 659         ipa/85449
 660         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
 661         recursion dependency to only apply to non-clones.
 662
 663 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 664
 665         ipa/85447
 666         * ipa-cp.c (create_specialized_node): Check that clones of
 667         self-recursive edges exist during IPA-CP.
 668
 669 2018-04-19  Toon Moene  <toon@moene.org>
 670
 671         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
 672         by -O3.
 673
 674 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 675
 676         PR tree-optimization/85467
 677         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
 678         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
 679         VECTOR_CST element to type.
 680
 681 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 682
 683         PR target/85397
 684         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
 685         * config/i386/i386.md (builtin_setjmp_setup): Removed.
 686         (builtin_longjmp): Likewise.
 687         (save_stack_nonlocal): New pattern.
 688         (restore_stack_nonlocal): Likewise.
 689
 690 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 691
 692         PR target/85404
 693         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 694         Replace ASM_OUTPUT_LABEL with fprintf.
 695
 696 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 697
 698         PR target/85417
 699         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 700         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
 701         * config/i386/i386-c.c (ix86_target_macros_internal): Also
 702         define __IBT__ and __SHSTK__ for -fcf-protection.
 703         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
 704         TARGET_IBT.
 705         (ix86_trampoline_init): Likewise.
 706         (x86_output_mi_thunk): Likewise.
 707         (ix86_notrack_prefixed_insn_p): Likewise.
 708         (ix86_option_override_internal): Don't disallow -fcf-protection.
 709         * config/i386/i386.md (rdssp<mode>): Also enable for
 710         -fcf-protection.
 711         (incssp<mode>): Likewise.
 712         (nop_endbr): Likewise.
 713         * config/i386/i386.opt (mcet): Change help message to built-in
 714         functions only.
 715         (mibt): Likewise.
 716         (mshstk): Likewise.
 717         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
 718         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
 719         enable CET built-in functions.
 720
 721 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
 722
 723         * common/config/i386/i386-common.c
 724         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
 725         OPTION_MASK_ISA_MOVDIRI_UNSET,
 726         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
 727         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
 728         * config.gcc (movdirintrin.h): New header.
 729         * config/i386/cpuid.h (bit_MOVDIRI,
 730         bit_MOVDIR64B): New bits.
 731         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
 732         and -mmvodir64b.
 733         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
 734         (VOID, PVOID, PCVOID)): New function types.
 735         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
 736         __builtin_ia32_directstoreu_u64,
 737          __builtin_ia32_movdir64b): New builtins.
 738         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
 739         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
 740         and -mmovdiri.
 741         (ix86_valid_target_attribute_inner_p): Ditto.
 742         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
 743         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
 744         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
 745         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
 746         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
 747         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
 748         (movdiri<mode>, movdir64b_<mode>): New.
 749         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
 750         * config/i386/immintrin.h: Include movdirintrin.h.
 751         * config/i386/movdirintrin.h: New file.
 752         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
 753
 754 2018-04-19  Richard Biener  <rguenther@suse.de>
 755
 756         PR middle-end/85455
 757         * cfg.c (clear_bb_flags): When loop state says we have
 758         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
 759
 760 2018-04-19  Richard Biener  <rguenther@suse.de>
 761
 762         PR tree-optimization/84737
 763         * tree-vect-data-refs.c (vect_copy_ref_info): New function
 764         copying restrict info.
 765         (vect_setup_realignment): Use it.
 766         * tree-vectorizer.h (vect_copy_ref_info): Declare.
 767         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
 768         the first DR to all generated stores.
 769         (vectorizable_load): Likewise for loads.
 770
 771 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 772
 773         PR tree-optimization/85446
 774         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
 775         the integral and pointer types to have the same precision.
 776
 777         * doc/install.texi: Document --disable-cet being the default and
 778         --enable-cet=auto.
 779
 780 2018-04-18  Martin Liska  <mliska@suse.cz>
 781
 782         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
 783         style.
 784
 785 2018-04-18  Martin Liska  <mliska@suse.cz>
 786
 787         Revert
 788         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
 789
 790         PR ipa/83983
 791         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
 792         arguments if they are comparable.
 793
 794 2018-04-18  Martin Liska  <mliska@suse.cz>
 795
 796         Revert
 797         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
 798
 799         PR lto/84805
 800         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
 801         incomplete types.
 802
 803 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
 804
 805         PR target/85388
 806         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
 807         ENDBR after calling __morestack.
 808
 809 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
 810
 811         PR jit/85384
 812         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
 813         by using gcc_base_ver to generate a gcc_driver_version, and use
 814         it when generating GCC_DRIVER_NAME.
 815         * configure: Regenerate.
 816
 817 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
 818
 819         PR target/81084
 820         * config.gcc: Obsolete powerpc*-*-*spe*.
 821
 822 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 823
 824         PR debug/84637
 825         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
 826         (stabstr_D): Change type of unum from unsigned int to
 827         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
 828         type.
 829
 830 2018-04-17  Jim Wilson  <jimw@sifive.com>
 831
 832         PR 84856
 833         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
 834         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
 835         Set arg_pointer_offset after using pretend_args_size.
 836
 837 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 838
 839         PR rtl-optimization/85431
 840         * dse.c (record_store): Ignore zero width stores.
 841
 842         PR sanitizer/85230
 843         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
 844         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
 845         __builtin_stack_restore rather than after it.
 846         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
 847         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
 848         argument instead of virtual_dynamic_stack_rtx.
 849
 850 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 851
 852         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
 853         New prototype.
 854         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
 855         Add note to error message to explain internal mapping of overloaded
 856         built-in function name to non-overloaded built-in function name.
 857         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
 858         function.
 859
 860 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
 861
 862         PR target/85424
 863         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
 864         where the inputs overlap with the output.
 865
 866 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 867
 868         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
 869         (=v, v) alternative and explicit "memory" attribute.
 870         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
 871         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 872         attributes.
 873         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
 874         "sselog1" type instead of "sselog".
 875         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
 876         "sselog".  Remove explicit "memory" attribute.
 877         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
 878         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 879         attributes.
 880         (vec_extract_hi_v32hi): Merge all alternatives into one, use
 881         "sselog1" type instead of "sselog".  Remove explicit "memory"
 882         attribute.
 883         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
 884         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 885         attribute.
 886         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
 887         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 888         attributes.
 889         (vec_extract_hi_v64qi): Merge all alternatives into one, use
 890         "sselog1" type instead of "sselog".  Remove explicit "memory"
 891         attribute.
 892         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
 893         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 894         attribute.
 895
 896         PR target/85430
 897         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
 898
 899         PR middle-end/85414
 900         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
 901         on a SUBREG.
 902
 903 2018-04-17  Martin Jambor  <mjambor@suse.cz>
 904
 905         PR ipa/85421
 906         * ipa-cp.c (create_specialized_node): Call
 907         expand_all_artificial_thunks if necessary.
 908
 909 2018-04-17  Martin Liska  <mliska@suse.cz>
 910
 911         PR lto/85405
 912         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
 913         in message, remote space in between '_G' and '('.
 914
 915 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 916
 917         PR target/85281
 918         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
 919         avx512f_vmcmp<mode>3<round_saeonly_name>,
 920         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
 921         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
 922         avx512f_rndscale<mode><round_saeonly_name>,
 923         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
 924         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
 925         Use %<iptr>2 instead of %2 for -masm=intel.
 926         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
 927         avx512f_vcvttss2usi<round_saeonly_name>,
 928         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
 929         -masm=intel.
 930         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
 931         avx512f_vcvttsd2usi<round_saeonly_name>,
 932         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
 933         Use %q1 instead of %1 for -masm=intel.
 934         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
 935         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
 936         of %3 for -masm=intel.
 937         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
 938         -masm=intel.
 939         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
 940         -masm=intel.
 941         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
 942         -masm=intel.
 943         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
 944         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
 945         %g1.
 946         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
 947         -masm=intel.
 948         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
 949         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
 950         %g1 and one with %0 and %1.
 951         (avx512er_vmrcp28<mode><round_saeonly_name>,
 952         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
 953         %1 for -masm=intel.
 954         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
 955         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
 956         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
 957         of %0 and %{%4%} for -masm=intel.
 958         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
 959         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
 960         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
 961         order of %0 and %{%5%}%{z%} for -masm=intel.
 962
 963 2018-04-17  Jan Hubicka  <jh@suse.cz>
 964
 965         PR lto/85405
 966         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
 967
 968 2018-04-17  Martin Liska  <mliska@suse.cz>
 969
 970         PR ipa/85329
 971         * multiple_target.c (create_dispatcher_calls): Set apostrophes
 972         for target_clone error message.  Make default implementation
 973         clone to be a local declaration.
 974         (separate_attrs): Add new argument and check for an empty
 975         string.
 976         (expand_target_clones): Handle it.
 977         (ipa_target_clone): Make redirection just for target_clones
 978         functions.
 979
 980 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
 981             Tom de Vries  <tom@codesourcery.com>
 982
 983         PR middle-end/84955
 984         * omp-expand.c (expand_oacc_for): Add dummy false branch for
 985         tiled basic blocks without omp continue statements.
 986
 987 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
 988
 989         PR target/83660
 990         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
 991         vec_extract expression as having side effects to make sure it gets
 992         a cleanup point.
 993
 994 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
 995
 996         PR target/85403
 997         * config/i386/i386.c (get_builtin_code_for_version): Check
 998         error_mark_node.
 999
1000 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
1001
1002         PR target/84331
1003         * gcc/config.gcc: Support "skylake".
1004         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
1005         PROCESSOR_SKYLAKE.
1006         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
1007         (processor_target_table): Add "skylake".
1008         (ix86_option_override_internal): Add "skylake".
1009         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
1010         PROCESSOR_CANNONLAKE.
1011         (get_builtin_code_for_version): Fix priority for
1012         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
1013         PROCESSOR_SKYLAKE-AVX512.
1014         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
1015         (processor_type): Add PROCESSOR_SKYLAKE.
1016
1017 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
1018             Jason Merrill  <jason@redhat.com>
1019
1020         PR c++/85112
1021         * convert.c (convert_to_integer_1): Use direct recursion for
1022         enumeral types and types with a precision less than the number
1023         of bits in their mode.
1024
1025 2018-04-16  Julia Koval  <julia.koval@intel.com>
1026
1027         PR target/84413
1028         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
1029         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
1030
1031 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
1032
1033         PR target/85293
1034         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
1035         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
1036         and -mno-direct-move.
1037
1038 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
1039
1040         PR target/83402
1041         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
1042         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
1043         Ensure negative shifts result in {0}.
1044
1045 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
1046
1047         PR rtl-optimization/79916
1048         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
1049         regs (if any) to define how to gnerate SD moves when LRA is in
1050         progress.
1051
1052 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
1053
1054         PR rtl-optimization/85393
1055         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
1056         * except.c (expand_dw2_landing_pad_for_region): Make static.
1057         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
1058         a label and unconditional jump to old_bb, rather than
1059         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
1060         basic block.
1061
1062         PR rtl-optimization/85376
1063         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
1064         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
1065         instead of a specific value.
1066
1067 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
1068             Bin Cheng  <bin.cheng@arm.com>
1069
1070         PR tree-optimization/82965
1071         PR tree-optimization/83991
1072         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
1073         by_profile_only parameter.
1074         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
1075         information if the loop was predicted to iterate too many times.
1076         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
1077
1078 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
1079
1080         PR lto/71991
1081         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
1082         always inline.
1083
1084 2018-04-13  Martin Liska  <mliska@suse.cz>
1085             Jakub Jelinek  <jakub@redhat.com>
1086
1087         PR middle-end/81657
1088         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
1089         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
1090         * builtins.c (expand_builtin_memory_copy_args): Use
1091         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
1092         handle dest_addr == pc_rtx.
1093
1094 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
1095
1096         PR target/85291
1097         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
1098         asked to not generate direct moves.
1099         (fix_trunc<mode>si2_stfiwx): Similar.
1100         (fix_trunc<mode>si2_internal): Similar.
1101
1102 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1103
1104         PR debug/83157
1105         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
1106         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
1107         lookup if dest in some wider mode is known to be const0_rtx and
1108         if so, record permanent equivalence for it to be ZERO_EXTEND of
1109         the narrower mode destination.
1110
1111 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
1112
1113         * lto-streamer-out.c (output_function): Revert 259346.
1114         * omp-expand.c (expand_oacc_for): Likewise.
1115
1116 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
1117
1118         PR rtl-optimization/85354
1119         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
1120         * sel-sched.c (sel_global_init): ... here.
1121
1122 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
1123
1124         PR target/85238
1125         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
1126         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
1127         mode for PE-COFF targets.
1128         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
1129         (i386_pe_asm_lto_end): Likewise.
1130         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
1131         (TARGET_ASM_LTO_END): Likewise.
1132         * config/i386/winnt.c (saved_debug_info_level): New static variable.
1133         (i386_pe_asm_lto_start): New function.
1134         (i386_pe_asm_lto_end): Likewise.
1135
1136 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
1137             Richard Biener  <rguenther@suse.de>
1138
1139         PR middle-end/84955
1140         * lto-streamer-out.c (output_function): Fix CFG loop state before
1141         streaming out.
1142         * omp-expand.c (expand_oacc_for): Handle calls to internal
1143         functions like regular functions.
1144
1145 2018-04-12  Richard Biener  <rguenther@suse.de>
1146
1147         PR lto/85371
1148         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
1149         for the early LTO debug to properly generate references to it
1150         during DIE emission.  Do not re-use that for the skeleton for
1151         split-dwarf.
1152         (dwarf2out_early_finish): Likewise.
1153
1154 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1155
1156         PR target/85328
1157         * config/i386/sse.md
1158         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
1159         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
1160         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
1161         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
1162         and output is a reg, avoid creating invalid lowpart subreg, but
1163         instead split into a 512-bit move.  Don't split if not AVX512VL,
1164         input is xmm16+ reg and output is a mem.
1165         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
1166         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
1167         xmm16+ reg and output is a mem.
1168
1169 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1170
1171         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
1172         also for flag_dwarf2_cfi_asm.
1173
1174 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1175
1176         PR rtl-optimization/85342
1177         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
1178         a bool scalar var inside of the loop instead.  Don't try to update
1179         recog_data.operand after failed apply_change_group.
1180
1181 2018-04-12  Tom de Vries  <tom@codesourcery.com>
1182
1183         PR target/85296
1184         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
1185         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
1186         array with flexible array member as array without given dimension.
1187         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
1188         argument for undefined param to true.
1189
1190 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
1191
1192         PR target/85321
1193         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
1194         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
1195         from PowerPC section.
1196         * config/rs6000/sysv4.opt (mcall-): Improve help text.
1197         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
1198         help text that is too long.
1199         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
1200         help text that is too long.
1201         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
1202         help text that is too long.
1203
1204 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
1205
1206         * config/alpha/alpha.md (stack_probe_internal): Rename
1207         from "probe_stack".  Update all callers.
1208
1209 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1210
1211         PR rtl-optimization/84566
1212         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
1213         sched_macro_fuse_insns.
1214
1215 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1216
1217         PR target/84301
1218         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
1219         (compute_block_dependences): ... from here.
1220
1221 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1222
1223         PR tree-optimization/85331
1224         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
1225         from int to HOST_WIDE_INT.
1226
1227 2018-04-11  Martin Jambor  <mjambor@suse.cz>
1228
1229         PR ipa/84149
1230         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
1231         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
1232         not the same as the source val.
1233         (cgraph_edge_brings_value_p): New parameter.
1234         (gather_edges_for_value): Pass destination value to
1235         cgraph_edge_brings_value_p.
1236         (perhaps_add_new_callers): Likewise.
1237         (get_info_about_necessary_edges): Likewise and exclude values brought
1238         only by self-recursive edges.
1239         (create_specialized_node): Redirect only clones of self-calling edges.
1240         (+self_recursive_pass_through_p): New function.
1241         (find_more_scalar_values_for_callers_subset): Use it.
1242         (find_aggregate_values_for_callers_subset): Likewise.
1243         (known_aggs_to_agg_replacement_list): Removed.
1244         (decide_whether_version_node): Re-calculate known constants for all
1245         remaining context clones.
1246
1247 2018-04-11  Richard Biener  <rguenther@suse.de>
1248
1249         PR lto/85339
1250         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
1251         from early DWARF output.
1252         (dwarf2out_early_finish): Output line info unconditionally into
1253         early DWARF and add reference to it.
1254
1255 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1256
1257         PR target/85281
1258         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
1259         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
1260         other than V2DFmode using iptr mode attribute.
1261         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
1262
1263 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1264
1265         PR rtl-optimization/84659
1266         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
1267
1268 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1269
1270         PR debug/85302
1271         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
1272         SIZEP is NULL.
1273         (output_loc_list): Pass address of a dummy size variable even in the
1274         locview handling loop.
1275         (index_location_lists): Add comment on why skip_loc_list_entry can't
1276         call size_of_locs.
1277
1278 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1279
1280         PR target/85261
1281         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
1282         into register.
1283
1284 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
1285
1286         PR target/85321
1287         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
1288         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
1289         and -mstring-compare-inline-limit.
1290
1291 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1292
1293         PR target/85287
1294         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
1295         for stack clash protection in a register whenever we need it to be in
1296         a register.
1297
1298 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1299
1300         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
1301         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
1302
1303 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1304
1305         PR target/85321
1306         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
1307         the help text.
1308         (mlong-double-): Ditto.
1309         * config/rs6000/sysv4.opt (msdata=): Ditto.
1310         (mtls-size=): Ditto.
1311
1312 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1313
1314         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
1315         erroneous entries for
1316         "vector int vec_ldl (int, long int *)", and
1317         "vector unsigned int vec_ldl (int, unsigned long int *)".
1318         Add comments and entries for
1319         "vector bool char vec_ldl (int, bool char *)",
1320         "vector bool short vec_ldl (int, bool short *)",
1321         "vector bool int vec_ldl (int, bool int *)",
1322         "vector bool long long vec_ldl (int, bool long long *)",
1323         "vector pixel vec_ldl (int, pixel *)",
1324         "vector long long vec_ldl (int, long long *)",
1325         "vector unsigned long long vec_ldl (int, unsigned long long *)".
1326         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
1327         type tree bool_long_long_type_node and correct definition of
1328         bool_V2DI_type_node to make reference to this new type tree.
1329         (rs6000_mangle_type): Replace erroneous reference to
1330         bool_long_type_node with bool_long_long_type_node.
1331         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
1332         comments to emphasize sign distinctions for char and int types and
1333         replace RS6000_BTI_bool_long constant with
1334         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
1335         use of RS6000_BTI_pixel.
1336         (bool_long_type_node): Remove this macro definition.
1337         (bool_long_long_type_node): New macro definition
1338
1339 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
1340
1341         PR rtl-optimization/85300
1342         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
1343         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
1344         simplify_unary_operation fails.
1345
1346 2018-04-10  Martin Liska  <mliska@suse.cz>
1347
1348         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
1349         cgraph_edge and ipa_ref.
1350
1351 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
1352
1353         PR target/85177
1354         PR target/85255
1355         * config/i386/sse.md
1356         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
1357         computation of the VEC_MERGE selector from mask.
1358         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
1359         Fix decoding of the VEC_MERGE selector into mask.
1360
1361 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
1362
1363         PR tree-optimization/85286
1364         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
1365
1366 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
1367
1368         * final.c (final_1): Set insn_last_address as well as
1369         insn_current_address.
1370
1371 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1372
1373         PR target/85173
1374         * explow.c (emit_stack_probe): Call validize_mem on memory location
1375         before passing it to gen_probe_stack.  Create address operand and
1376         legitimize it for the probe_stack_address case.
1377
1378 2018-04-09  Jan Hubicka  <jh@suse.cz>
1379
1380         PR lto/85078
1381         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
1382         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
1383         * tree.c (free_lang_data_in_type): Fix handling of binfos;
1384         walk basetypes.
1385         (free_lang_data): Rebuild type inheritance graph.
1386
1387 2018-04-09  Martin Sebor  <msebor@redhat.com>
1388
1389         * invoke.texi (-finline-small-functions): Mention other optimization
1390         options.
1391         (-findirect-inlining, -fpartial-inlining): Same.
1392         (-finline-functions-called-once): Same.
1393         (-freorder-blocks-and-partition): Same.
1394
1395 2018-04-09  Jan Hubicka  <jh@suse.cz>
1396
1397         PR rtl/84058
1398         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
1399         jumps; choose last target that matches the criteria (i.e.
1400         no partition changes for non-crossing jumps).
1401         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
1402         support for redirecting crossing jumps to non-crossing.
1403
1404 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
1405
1406         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
1407         also for naked functions.
1408
1409 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
1410
1411         * config/arc/arc.md (add_shift): New pattern.
1412         (add_shift2): Likewise.
1413         (sub_shift): Likewise.
1414         (sub_shift_cmp0_noout): Likewise.
1415         (compare_si_ashiftsi): Likewise.
1416         (xbfu_cmp0_noout): New combine pattern.
1417         (xbfu_cmp0"): Likewise.
1418         (movsi_set_cc_insn): Place the predicable variant first.
1419         (commutative_binary_cmp0_noout): Remove clobber.
1420         (commutative_binary_cmp0): New pattern.
1421         (noncommutative_binary_cmp0): Likewise.
1422         (noncommutative_binary_cmp0_noout): Likewise.
1423         (noncommutative_binary_comparison_result_used): Removed.
1424         (rsub_cmp0): New pattern.
1425         (rsub_cmp0_noout): Likewise.
1426         (extzvsi): Changed, keep only meaningful variants.
1427         (SQH, SEZ): New iterators.
1428         (SQH_postfix): New mode attribute.
1429         (SEZ_prefix): New code attribute.
1430         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
1431         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
1432         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
1433         of numerical value.
1434         (noncommutative_operator): Check the availability of barrel
1435         shifter option.
1436
1437 2018-04-09  Richard Biener  <rguenther@suse.de>
1438
1439         PR tree-optimization/85284
1440         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
1441         Only use the niter constraining form of simple_iv when the exit
1442         is always executed.
1443
1444 2018-04-09  Tom de Vries  <tom@codesourcery.com>
1445
1446         PR target/84041
1447         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
1448         (define_expand "*memory_barrier"): New define_expand.
1449         (define_insn "memory_barrier"): New insn.
1450
1451 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1452
1453         PR rtl-optimization/80463
1454         PR rtl-optimization/83972
1455         PR rtl-optimization/83480
1456
1457         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
1458         correct producer for the insn.
1459         (tidy_control_flow): Fixup seqnos in case of debug insns.
1460
1461 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1462
1463         PR rtl-optimization/83913
1464
1465         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
1466         different sched-times when merging exprs.
1467
1468 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1469
1470         PR rtl-optimization/83962
1471
1472         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
1473         tidy_fallthru_edge and tidy_control_flow.
1474
1475 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1476
1477         PR rtl-optimization/83530
1478
1479         * sel-sched.c (force_next_insn): New global variable.
1480         (remove_insn_for_debug): When force_next_insn is true, also leave only
1481         next insn in the ready list.
1482         (sel_sched_region): When the region wasn't scheduled, make another pass
1483         over it with force_next_insn set to 1.
1484
1485 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1486
1487         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
1488         into tm_file.
1489         * config/nds32/constants.md (unspec_volatile_element): Add enum values
1490         for interrupt control.
1491         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1492         functions for interrupt control.
1493         * config/nds32/nds32-intrinsic.md: Likewise.
1494         * config/nds32/nds32_intrinsic.h: Likewise.
1495         * config/nds32/nds32.h (nds32_builtins): Likewise.
1496
1497 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1498
1499         * config/nds32/nds32.c (nds32_init_machine_status,
1500         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
1501         strict_aligned_p field.
1502         (nds32_expand_to_rtl_hook): New function.
1503         (TARGET_EXPAND_TO_RTL_HOOK): Define.
1504         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
1505
1506 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1507             Chung-Ju Wu  <jasonwucj@gmail.com>
1508
1509         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
1510         * config/nds32/nds32-n7.md: New file.
1511         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
1512         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
1513         pipeline.
1514         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
1515         * config/nds32/nds32.md (pipeline_model): Add n7.
1516         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
1517         * config/nds32/pipelines.md: Include n7 settings.
1518
1519 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1520             Chung-Ju Wu  <jasonwucj@gmail.com>
1521
1522         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
1523         * config/nds32/nds32-e8.md: New file.
1524         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
1525         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
1526         pipeline.
1527         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
1528         * config/nds32/nds32.md (pipeline_model): Add e8.
1529         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
1530         * config/nds32/pipelines.md: Include e8 settings.
1531
1532 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1533             Chung-Ju Wu  <jasonwucj@gmail.com>
1534
1535         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
1536         * config/nds32/nds32-n8.md: New file.
1537         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
1538         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
1539         pipeline.
1540         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
1541         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
1542         * config/nds32/nds32.md (pipeline_model): Add n8.
1543         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
1544         * config/nds32/pipelines.md: Include n8 settings.
1545
1546 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1547             Chung-Ju Wu  <jasonwucj@gmail.com>
1548
1549         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
1550         * config/nds32/nds32-n9-2r1w.md: New file.
1551         * config/nds32/nds32-n9-3r2w.md: New file.
1552         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
1553         nds32_register_ports): New or modify for cpu n9.
1554         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
1555         pipeline.
1556         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
1557         * config/nds32/nds32-utils.c: New file.
1558         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
1559         TARGET_MUL_SLOW): Define.
1560         * config/nds32/nds32.md (pipeline_model): New attribute.
1561         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
1562         New options that support cpu n9.
1563         * config/nds32/pipelines.md: Include n9 settings.
1564         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
1565
1566 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1567
1568         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
1569         information if necessary.
1570         (output_cond_branch_compare_zero): Likewise.
1571         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
1572         (nds32_target_alignment): Refine for alignment.
1573         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
1574         (FUNCTION_BOUNDARY): Modify.
1575         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
1576         align case.
1577         * config/nds32/nds32.opt (malways-align, malign-functions): New.
1578
1579 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1580
1581         * config/nds32/constants.md (unspec_volatile_element): Add values for
1582         TLB operation and data prefetch.
1583         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1584         functions for TLB operation and data prefetch.
1585         * config/nds32/nds32-intrinsic.md: Likewise.
1586         * config/nds32/nds32_intrinsic.h: Likewise.
1587         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
1588         (nds32_print_operand): Likewise.
1589         * config/nds32/nds32.h (nds32_builtins): Likewise.
1590
1591 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
1592         Andrew Pinski <pinsika@gcc.gnu.org>
1593
1594         PR middle-end/82976
1595         * match.pd: Use constant_boolean_node of correct type instead of
1596         boolean_true_node or boolean_false_node for simplifying
1597         pointer comparisons to zero.
1598
1599 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1600
1601         PR tree-optimization/80021
1602         * tree.c (verify_type_variant): Make error call in verify_variant_match
1603         translatable and remove final full stop.
1604
1605 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1606
1607         * config/nds32/constants.md (unspec_volatile_element): Add
1608         UNSPEC_VOLATILE_EH_RETURN.
1609         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
1610         nds32_output_stack_pop): Support dwarf exception handling process.
1611         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
1612         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
1613         exception handling process.
1614         (nds32_compute_stack_frame): Likewise.
1615         (nds32_return_addr_rtx): Likewise.
1616         (nds32_initial_elimination_offset): Likewise.
1617         (nds32_expand_prologue): Likewise.
1618         (nds32_expand_epilogue): Likewise.
1619         (nds32_dynamic_chain_address): New function.
1620         * config/nds32/nds32.h (machine_function): Add fields for dwarf
1621         exception handling.
1622         (DYNAMIC_CHAIN_ADDRESS): Define.
1623         (EH_RETURN_DATA_REGNO): Define.
1624         (EH_RETURN_STACKADJ_RTX): Define.
1625         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
1626         patterns for dwarf exception handling.
1627
1628 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1629
1630         * config/nds32/nds32.h: Clean up obsolete macros.
1631
1632 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1633
1634         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1635         Add enum values for particular instructions.
1636         * config/nds32/nds32-intrinsic.c: Implementation of expanding
1637         particular intrinsic functions.
1638         * config/nds32/nds32-intrinsic.md: Likewise.
1639         * config/nds32/nds32_intrinsic.h: Likewise.
1640         * config/nds32/nds32.h (nds32_builtins): Likewise.
1641         * config/nds32/nds32.md (type): Add pbsad and pbsada.
1642         (btst, ave): New patterns for particular instructions.
1643
1644 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1645
1646         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1647         Add enum values for atomic load/store and memory sync.
1648         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
1649         and memory sync.
1650         * config/nds32/nds32-intrinsic.md: Likewise.
1651         * config/nds32/nds32_intrinsic.h: Likewise.
1652         * config/nds32/nds32.h (nds32_builtins): Likewise.
1653
1654 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1655
1656         PR tree-optimization/85257
1657         * fold-const.c (native_encode_vector): If not all elts could fit
1658         and off is -1, return 0 rather than offset.
1659         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
1660         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
1661         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
1662         adjust buffer in native_interpret_expr call.
1663
1664 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1665
1666         * config/nds32/constants.md (unspec_volatile_element): Add cache
1667         control enum values.
1668         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
1669         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
1670         * config/nds32/nds32.c (nds32_cctl_names): New.
1671         (nds32_print_operand): Handle cache control register names.
1672         * config/nds32/nds32.h (nds32_builtins): New enum values.
1673         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
1674         macros.
1675         * config/nds32/nds32.md (type): Add mmu.
1676         * config/nds32/pipelines.md (simple_insn): Add mmu.
1677
1678 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1679
1680         * config/nds32/nds32.md (type): Remove call.
1681         * config/nds32/pipelines.md (simple_insn): Likewise.
1682
1683 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1684
1685         * config/nds32/constants.md (unspec_volatile_element): Add
1686         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
1687         UNSPEC_VOLATILE_FMFCFG.
1688         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
1689         description for fmfcfg and fmfcsr.
1690         (bdesc_1arg): Add fmtcsr.
1691         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
1692         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
1693         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
1694         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
1695         unspec_fmfcfg): New patterns.
1696         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
1697         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
1698         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
1699         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
1700         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
1701         __nds32__fmfcfg): Define.
1702
1703 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1704
1705         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
1706         intrinsic register names.
1707         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
1708         intrinsic register enum values and macros.
1709
1710 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1711
1712         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
1713         for load/store addressing form.
1714         (nds32_print_operand_address): Likewise.
1715
1716 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
1717
1718         PR target/85196
1719         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
1720         based on LABEL_REF.  Remove useless assertion.
1721         (pic_address_needs_scratch): Fix formatting.
1722         (sparc_legitimize_pic_address): Minor tweaks.
1723         (sparc_delegitimize_address): Adjust assertion accordingly.
1724         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
1725         into symbolic_operand.
1726         (movsi_high_pic_label_ref): Likewise.
1727         (movsi_lo_sum_pic_label_ref): Likewise.
1728         (movdi_pic_label_ref): Likewise.
1729         (movdi_high_pic_label_ref): Likewise.
1730         (movdi_lo_sum_pic_label_ref): Likewise.
1731
1732 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
1733
1734         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
1735         custom LIB_SPEC setup.
1736
1737 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
1738             Kito Cheng  <kito.cheng@gmail.com>
1739
1740         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
1741         * config/riscv/freebsd.h: New.
1742
1743 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1744
1745         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
1746         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
1747         file.
1748
1749 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1750             Kito Cheng  <kito.cheng@gmail.com>
1751
1752         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
1753         nds32_output_call, nds32_symbol_binds_local_p): New functions.
1754         * config/nds32/nds32-protos.h (nds32_output_call,
1755         nds32_output_return): Declare.
1756         * config/nds32/nds32.md: Refine all the call and return patterns.
1757
1758 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1759
1760         PR debug/85252
1761         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
1762         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
1763
1764         PR rtl-optimization/84872
1765         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
1766         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
1767         EDGE_CROSSING edge.
1768
1769 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
1770
1771         * expr.c (copy_blkmode_to_reg): Revert 254862.
1772         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
1773
1774 2018-04-06  Richard Biener  <rguenther@suse.de>
1775
1776         PR middle-end/85244
1777         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
1778         after seeing a component reference with an adjacent field.  Treat
1779         refs to arrays at struct end of external decls similar to
1780         refs to unconstrained commons.
1781
1782 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1783
1784         PR sanitizer/85213
1785         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
1786         look through SAVE_EXPRs with non-side-effects argument.  Adjust
1787         recursive calls.
1788         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
1789         save_p here.
1790
1791 2018-04-06  Richard Biener  <rguenther@suse.de>
1792
1793         PR middle-end/85180
1794         * alias.c (find_base_term): New wrapper around find_base_term
1795         unwinding CSELIB_VAL_PTR changes.
1796         (find_base_term): Do not restore CSELIB_VAL_PTR during the
1797         recursion.
1798
1799 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1800
1801         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
1802         instructions.
1803         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
1804         constant definitions.
1805         ("nop"): lr 0,0 -> nopr r0
1806         ("nop_lr0", "nop_lr1"): New insn definitions.
1807
1808 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1809
1810         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
1811         NDS32_V3PUSH_AVAILABLE_P macro.
1812
1813 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
1814             Chung-Ju Wu  <jasonwucj@gmail.com>
1815
1816         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
1817         (nds32*-*-*): Add float and fpu_config into supported_defaults.
1818         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
1819         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
1820         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
1821         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
1822         * config/nds32/constraints.md: New constraints and checking for hard
1823         float configuration.
1824         * config/nds32/iterators.md: New mode iterator and attribute for hard
1825         float configuration.
1826         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
1827         patterns.
1828         * config/nds32/nds32-fpu.md: New file.
1829         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
1830         deal with hard float code generation.
1831         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
1832         ARCH_V3S.
1833         (abi_type, float_reg_number): New enum type.
1834         * config/nds32/nds32-predicates.c: New predicates for hard float.
1835         * config/nds32/nds32-protos.h: Declare functions for hard float.
1836         * config/nds32/nds32.c: Implementation for hard float configuration.
1837         * config/nds32/nds32.h: Definitions for hard float configuration.
1838         * config/nds32/nds32.md: Include hard float machine description and
1839         modify patterns for hard float configuration.
1840         * config/nds32/nds32.opt: New options for hard float configuration.
1841         * config/nds32/predicates.md: New predicates for hard float
1842         configuration.
1843
1844 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
1845
1846         * common/config/nds32/nds32-common.c
1847         (nds32_option_optimization_table): Enable -mreleax-hint by default.
1848
1849 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
1850
1851         PR middle-end/85195
1852         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
1853         CONSTRUCTOR_ELT (ctor, ...)->value.
1854
1855 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
1856
1857         PR target/85193
1858         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
1859
1860 2018-04-05  Tom de Vries  <tom@codesourcery.com>
1861
1862         PR target/85204
1863         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
1864         cond jump.
1865
1866 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1867             Kito Cheng  <kito.cheng@gmail.com>
1868
1869         * config/nds32/constraints.md (U33): Fine-tune checking condition.
1870         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
1871         * config/nds32/nds32.h (nds32_16bit_address_type): Add
1872         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
1873
1874 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1875             Kito Cheng  <kito.cheng@gmail.com>
1876
1877         * config/nds32/constraints.md (Ufe): New memory constraint.
1878         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
1879         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
1880         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
1881         operands.
1882         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
1883         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
1884
1885 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1886
1887         * config/nds32/nds32.md: Use optimize_size in the condition for
1888         alu-shift instructions.
1889
1890 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1891
1892         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
1893
1894 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1895
1896         * config/nds32/nds32.md (negsi2): Refine pattern.
1897
1898 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1899             Chung-Ju Wu  <jasonwucj@gmail.com>
1900
1901         * config/nds32/iterators.md (shift_rotate): New code iterator.
1902         (shift): New code attribute.
1903         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
1904         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
1905         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
1906         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
1907         bit-wise operations.
1908         (andsi3, *andsi3): Ditto.
1909         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
1910         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
1911         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
1912         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
1913         nds32_ior_operand, nds32_xor_operand): New predicates.
1914
1915 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1916
1917         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
1918         (addsi3, subsi3): ... this.
1919
1920 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1921
1922         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
1923
1924 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1925
1926         * config/nds32/nds32.md: Adjust indention.
1927
1928 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1929
1930         * config/nds32/nds32.md (feature): New attribute.
1931
1932 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1933
1934         * config/nds32/nds32.md (subtype): New attribute.
1935
1936 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1937
1938         PR target/85203
1939         * config/arm/arm-builtins.c (arm_expand_builtin): Change
1940         expansion to perform a bitwise AND of the argument followed by a
1941         boolean negation of the result.
1942
1943 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
1944
1945         PR rtl-optimization/84878
1946         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
1947         the basic block.  Assert the use reference is not artificial and that
1948         it has an associated insn.
1949
1950 2018-04-04  Michael Matz  <matz@suse.de>
1951
1952         * builtins.c (compute_objsize): Pass correct operand
1953         to array_at_struct_end_p.
1954
1955 2018-04-04  Richard Biener  <rguenther@suse.de>
1956
1957         PR lto/85176
1958         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
1959         from contexts for DINFO_LEVEL_TERSE and below.
1960
1961 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1962
1963         * config/nds32/nds32-doubleword.md (move_<mode>): Require
1964         resiter_operand condition.
1965         * config/nds32/nds32.md (*move<mode>): Ditto.
1966
1967 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1968             Monk Chiang  <sh.chiang04@gmail.com>
1969
1970         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
1971
1972 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1973
1974         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
1975
1976 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1977             Kito Cheng  <kito.cheng@gmail.com>
1978
1979         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
1980         nds32_cond_code_str, output_cond_branch,
1981         output_cond_branch_compare_zero, nds32_expand_cbranch,
1982         nds32_expand_cstore, nds32_expand_movcc,
1983         nds32_output_cbranchsi4_equality_zero,
1984         nds32_output_cbranchsi4_equality_reg,
1985         nds32_output_cbranchsi4_equality_reg_or_const_int,
1986         nds32_output_cbranchsi4_greater_less_zero: New functions.
1987         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
1988         nds32_expand_cstore, nds32_expand_movcc,
1989         nds32_output_cbranchsi4_equality_zero,
1990         nds32_output_cbranchsi4_equality_reg,
1991         nds32_output_cbranchsi4_equality_reg_or_const_int,
1992         nds32_output_cbranchsi4_greater_less_zero): Declare.
1993         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
1994         nds32_rimm11s_operand): New predicates.
1995         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
1996         * config/nds32/nds32.md: Rewrite all the branch and conditional move
1997         patterns.
1998
1999 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
2000
2001         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
2002         * config/nds32/nds32.md: Ditto.
2003         * config/nds32/pipelines.md: Ditto.
2004
2005 2018-04-04  Richard Biener  <rguenther@suse.de>
2006
2007         PR tree-optimization/85168
2008         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
2009         propagating abnormals.
2010
2011 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2012
2013         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
2014
2015 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2016             Kito Cheng  <kito.cheng@gmail.com>
2017
2018         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
2019         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
2020         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
2021         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
2022         * config/nds32/nds32.md (sibcall_internal): New.
2023         (sibcall_register): Remove.
2024         (sibcall_immediate): Remove.
2025         (sibcall_value_internal): New.
2026         (sibcall_value_register): Remove.
2027         (sibcall_value_immediate): Remove.
2028         * config/nds32/predicates.md (nds32_general_register_operand): New.
2029         (nds32_call_address_operand): New.
2030
2031 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
2032
2033         PR rtl-optimization/85167
2034         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
2035         bb_defs if *split_p, instead preinitialize it to NULL.
2036
2037         PR tree-optimization/85156
2038         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
2039         evaluating the argument multiple times.
2040
2041 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
2042
2043         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
2044         than vector.
2045         (_mm_cvtpd_ps): Likewise.
2046         (_mm_cvttpd_epi32): Likewise.
2047         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
2048         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
2049         vector, pixel, and bool following altivec.h include.
2050
2051 2018-04-03  Martin Sebor  <msebor@redhat.com>
2052
2053         * doc/extend.texi (Common Function Attributes): Clarify.
2054         (const attribute): Likewise.
2055         (pure attribute): Likewise.
2056
2057 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
2058
2059         PR target/85169
2060         * config/i386/i386.c (ix86_expand_vector_set): Use
2061         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
2062
2063 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
2064
2065         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
2066         instructions when changing rounding bits to preserve precision bits
2067         in the x87 control word.
2068
2069 2018-04-03  Martin Liska  <mliska@suse.cz>
2070
2071         PR tree-optimization/82491
2072         * rtl.h (strip_offset_and_add): Replace += suboffset with
2073         poly_uint64 () + suboffset.
2074
2075 2018-03-29  Martin Liska  <mliska@suse.cz>
2076             Martin Jambor  <mjambor@suse.cz>
2077
2078         PR ipa/84947
2079         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
2080         param_type is not an integral or pointer type.
2081
2082 2018-04-03  Richard Biener  <rguenther@suse.de>
2083
2084         * sese.h (recompute_all_dominators): Remove.
2085
2086 2018-04-02  Martin Sebor  <msebor@redhat.com>
2087
2088         * doc/invoke.texi (-Wrestrict): Fix typos.
2089
2090 2018-04-02  Jim Wilson  <jimw@sifive.com>
2091
2092         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
2093         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
2094         (<optab>di3, <optab>si3_extend): Likewise.
2095         (<optab>si3_mask, <optab>si3_mask_1): New.
2096         (<optab>di3_mask, <optab>di3_mask_1): New.
2097         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
2098         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
2099         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
2100
2101 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
2102
2103         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
2104         example.
2105
2106 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
2107
2108         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
2109         (nds32_canonicalize_comparison): New function.
2110
2111 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2112             Kito Cheng  <kito.cheng@gmail.com>
2113             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
2114
2115         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
2116         * config/nds32/constants.md (unspec_volatile_element): Add
2117         UNSPEC_VOLATILE_RELAX_GROUP.
2118         * config/nds32/nds32-relax-opt.c: New file.
2119         * config/nds32/nds32-predicates.c
2120         (nds32_symbol_load_store_p): New function.
2121         * config/nds32/nds32-protos.h
2122         (nds32_symbol_load_store_p): Declare function.
2123         (make_pass_nds32_relax_opt): Declare new rtl pass function.
2124         * config/nds32/nds32.c
2125         (nds32_register_pass): New function to register pass.
2126         (nds32_register_passes): New function to register passes.
2127         * config/nds32/nds32.md (relax_group): New pattern.
2128         * config/nds32/nds32.opt (mrelax-hint): New option.
2129         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
2130
2131 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
2132
2133         * config/nds32/t-nds32: Modify files dependency.
2134
2135 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2136
2137         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
2138         (PROFILE_HOOK): Define its implementation.
2139
2140 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2141
2142         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
2143         type and 32-bit size.
2144
2145 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
2146
2147         PR middle-end/85090
2148         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
2149         (V_128_256): New mode iterator.
2150         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
2151         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
2152         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
2153         of V.
2154         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
2155         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
2156
2157 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
2158
2159         PR target/83315
2160         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
2161         NaN inputs correctly.
2162
2163 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
2164
2165         PR target/80546
2166         * config/rs6000/vsx.md (??r): New mode attribute.
2167         (*vsx_mov<mode>_64bit): Use it.
2168         (*vsx_mov<mode>_32bit): Likewise.
2169
2170 2018-03-30  Martin Sebor  <msebor@redhat.com>
2171
2172         PR tree-optimization/84818
2173         * builtins.c (check_access): Use warning_n.
2174
2175 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
2176
2177         PR target/83822
2178         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
2179         condition.
2180         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
2181         condition.
2182
2183 2018-03-30  Julia Koval  <julia.koval@intel.com>
2184
2185         PR target/84413
2186         * x86-tune.def (movx, partial_reg_dependency): Enable for
2187         m_SKYLAKE_AVX512.
2188
2189 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
2190
2191         PR inline-asm/84985
2192         * lra-constraints.c (process_alt_operands): Move setting
2193         this_alternative_matches below.
2194
2195 2018-03-29  Martin Liska  <mliska@suse.cz>
2196
2197         PR lto/84995.
2198         * doc/invoke.texi: Document how LTO works with debug info.
2199         Describe auto-load support of binutils.  Mention 'x86-64'
2200         as valid option value of -march option.
2201
2202 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
2203
2204         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
2205
2206         PR c/85094
2207         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
2208         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
2209         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
2210         checking.
2211
2212 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
2213
2214         PR target/84912
2215         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
2216         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
2217         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
2218         for RS6000_BTM_POWERPC64.
2219         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
2220         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
2221         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
2222         definition.
2223         (DIVDE): Use it.
2224         (DIVDEU): Likewise.
2225
2226 2018-03-28 Carl Love  <cel@us.ibm.com>
2227
2228         Revert
2229         2017-09-27  Carl Love  <cel@us.ibm.com>
2230
2231         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
2232         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
2233         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
2234         fctiw instruction.
2235
2236 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
2237
2238         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
2239         instead of __vector bool.
2240         (_mm_max_pu8): Likewise.
2241         (_mm_min_pi16): Likewise.
2242
2243 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
2244
2245         PR target/84912
2246         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
2247         (DIVWEUO): Likewise.
2248         (DIVDEO): Likewise.
2249         (DIVDEUO): Likewise.
2250         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
2251         DIVWEUO and DIVDEUO.
2252         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
2253         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
2254         (div_extend): Likewise.
2255         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
2256         builtin function.
2257         (__builtin_divweuo): Likewise.
2258         (__builtin_divdeo): Likewise.
2259         (__builtin_divdeuo): Likewise.
2260
2261 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
2262
2263         PR target/85095
2264         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
2265         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
2266
2267         PR tree-optimization/82004
2268         * gimple-match-head.c (optimize_pow_to_exp): New function.
2269         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
2270         Don't fold to exp if optimize_pow_to_exp is false.
2271
2272 2018-03-28  Martin Liska  <mliska@suse.cz>
2273
2274         PR other/84819
2275         * calls.c (initialize_argument_information): Fix trailing space.
2276         * common.opt: Fix typo and provide better explanation for
2277         -fsanitize-coverage option.
2278         * config/i386/i386.opt: Fix typo.
2279
2280 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
2281             Martin Liska  <mliska@suse.cz>
2282
2283         PR sanitizer/85081
2284         * gimplify.c (asan_poison_variable): Don't do the check for
2285         gimplify_omp_ctxp here.
2286         (gimplify_decl_expr): Do it here.
2287         (gimplify_target_expr): Likewise.
2288
2289 2018-03-28  Martin Liska  <mliska@suse.cz>
2290
2291         PR target/84988
2292         * config/i386/i386.c (ix86_function_arg_advance): Do not call
2293         chkp_type_bounds_count if MPX is not enabled.
2294
2295 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
2296
2297         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
2298
2299 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
2300
2301         PR target/84914
2302         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
2303         function to create the function decl for complex long double
2304         multiply and divide for -mabi=ieeelongdouble.
2305         (init_float128_ieee): Call it.
2306
2307 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
2308
2309         PR target/85044
2310         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
2311         -fcf-protection=branch -mibt.
2312         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
2313
2314 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2315
2316         PR target/81863
2317         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
2318
2319 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
2320
2321         PR target/85056
2322         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
2323         extern array declarations.
2324
2325 2018-03-27  Richard Biener  <rguenther@suse.de>
2326
2327         PR middle-end/84067
2328         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
2329         explicit single_use checks.
2330
2331 2018-03-27  Richard Biener  <rguenther@suse.de>
2332
2333         PR tree-optimization/85082
2334         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
2335         Valueize the VUSE.
2336
2337 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2338
2339         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
2340         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
2341         Turn on fasynchronous-unwind-tables and funwind-tables.
2342
2343 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
2344
2345         PR target/85073
2346         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
2347         (*bmi_blsr_<mode>_ccz): Ditto.
2348
2349 2018-03-26  Tom de Vries  <tom@codesourcery.com>
2350
2351         PR tree-optimization/85063
2352         * omp-general.c (offloading_function_p): New function.  Factor out
2353         of ...
2354         * omp-offload.c (pass_omp_target_link::gate): ... here.
2355         * omp-general.h (offloading_function_p): Declare.
2356         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
2357         with attribute omp declare target for offloading functions.
2358
2359 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
2360
2361         PR tree-optimization/84005
2362         * tree-data-ref.h (get_base_for_alignment): Declare.
2363         * tree-data-ref.c (get_base_for_alignment_1): New function.
2364         (get_base_for_alignment): Likewise.
2365         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
2366         get_base_for_alignment to find a suitable base object, instead
2367         of always using drb->base_address.
2368
2369 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
2370
2371         PR inline-asm/85022
2372         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
2373         known size by default.
2374
2375 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
2376
2377         PR inline-asm/85030
2378         * lra-constraints.c (process_alt_operands): Don't match BLKmode
2379         and non BLKmode operands.
2380
2381 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2382
2383         PR target/85026
2384         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
2385         Clean up attributes.
2386
2387 2018-03-23  Richard Biener  <rguenther@suse.de>
2388
2389         PR debug/85020
2390         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
2391         we are going to emit early debug for LTO.
2392
2393 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
2394
2395         PR inline-asm/85034
2396         * function.c (match_asm_constraints_1): Don't optimize if input
2397         doesn't satisfy general_operand predicate for output's mode.
2398
2399         PR inline-asm/85022
2400         * alias.c (write_dependence_p): Don't require for x_canonicalized
2401         non-VOIDmode if x has VOIDmode.
2402
2403         PR sanitizer/85029
2404         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
2405         just don't try to optimize it rather than assert it never happens.
2406
2407 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2408
2409         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
2410         macro expansions for definition of ST_INTERNAL_<mode> and
2411         LD_INTERNAL_<mode> builtins.
2412         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
2413         Remove prototype.
2414         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
2415         function.
2416         (altivec_expand_st_builtin): Likewise.
2417         (altivec_expand_builtin): Remove calls to deleted functions.
2418         (rs6000_address_for_altivec): Delete this function.
2419         * config/rs6000/vector.md: Remove expands for
2420         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
2421
2422 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
2423
2424         PR target/84826
2425         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
2426         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
2427         re-computing once computed.
2428         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
2429         (arm_init_machine_status): Initialize
2430         machine->static_chain_stack_bytes.
2431
2432 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2433
2434         PR target/84760
2435         * doc/extend.texi: Add four new prototypes for vec_ld.
2436         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
2437         definitions for more logical presentation.
2438         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
2439         entries for V1TI variants of __builtin_altivec_ld builtin.
2440         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
2441         handling of V1TI variant of LVX icode pattern.
2442         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
2443         (rs6000_gimple_fold_builtin): Likewise.
2444         (altivec_init_builtins): Add code to define
2445         __builtin_altivec_lvx_v1ti function.
2446
2447 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2448
2449         PR inline-asm/84941
2450         * function.c (match_asm_constraints_1): Don't do the optimization
2451         if input isn't a REG, SUBREG, MEM or constant.
2452
2453 2018-03-22  Tom de Vries  <tom@codesourcery.com>
2454
2455         PR tree-optimization/84956
2456         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
2457         bb_has_abnormal_pred.
2458
2459 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2460
2461         PR sanitizer/85018
2462         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
2463         DECL_INITIAL (decl) to decl at the end.
2464         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
2465         adjust the comment.
2466
2467 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
2468
2469         * doc/extend.texi (__builtin_tgmath): Document when complex
2470         integer types are treated as _Complex _Float64.
2471
2472 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2473
2474         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
2475
2476 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2477
2478         PR tree-optimization/84960
2479         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
2480         if it is ENTRY block, move them into single succ of ENTRY in that case.
2481
2482 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
2483
2484         PR tree-optimization/84811
2485         * poly-int.h (poly_span_traits): Remove the T3 parameter and
2486         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
2487         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
2488         (known_subrange_p): Update accordingly.  Cast each value involved
2489         in the size comparison, rather than casting the result of the
2490         subtraction.
2491
2492 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2493
2494         PR tree-optimization/84982
2495         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
2496         by flipping the least significant bit rather than all bits from
2497         bitpos to bitpos + bitsize - 1.
2498
2499 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2500
2501         * doc/extend.texi (Deprecated Features): Remove mention of
2502         long-deleted deprecations.
2503
2504 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
2505
2506         PR jit/84288
2507         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
2508         * configure: Regenerate.
2509
2510 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2511
2512         PR tree-optimization/83126
2513         * tree-parloops.c (num_phis): New function.
2514         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
2515
2516 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2517
2518         * doc/extend.texi (Deprecated Features): Update deprecated flags,
2519         mention anon-struct/union members and trailing attributes.
2520
2521 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
2522
2523         PR tree-optimization/84969
2524         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
2525         builtin memset partitions if they set different rhs values.
2526
2527 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2528
2529         PR rtl-optimization/84989
2530         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
2531         VEC_DUPLICATE with scalar result mode.
2532
2533 2018-03-21  Martin Liska  <mliska@suse.cz>
2534
2535         PR ipa/84963
2536         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
2537         not intended return statement.
2538
2539 2018-03-21  Martin Liska  <mliska@suse.cz>
2540
2541         PR target/84988
2542         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
2543         (chkp_find_bound_slots_1): Limit number of iterations.
2544
2545 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
2546
2547         PR target/84838
2548         * Minor grammar fixes for x86 options.
2549
2550 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2551
2552         PR debug/84875
2553         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
2554         holding REG_CFA_RESTORE notes, instead turn them into a USE.
2555
2556 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
2557
2558         PR target/83789
2559         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
2560         (altivec_lvx_<mode>_1op): Likewise.
2561         (altivec_stvx_<mode>_2op): Likewise.
2562         (altivec_stvx_<mode>_1op): Likewise.
2563         (altivec_lvx_<VM2:mode>): New define_expand.
2564         (altivec_stvx_<VM2:mode>): Likewise.
2565         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
2566         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2567         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
2568         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2569         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
2570         (rs6000_gen_lvx): Likewise.
2571         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
2572         (altivec_expand_stv_builtin): Likewise.
2573         (altivec_expand_builtin): Likewise.
2574         * config/rs6000/vector.md: Likewise.
2575
2576 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2577
2578         PR target/82518
2579         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
2580         BYTES_BIG_ENDIAN.
2581
2582 2018-03-20  Richard Biener  <rguenther@suse.de>
2583
2584         PR target/84986
2585         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
2586         sign-conversions as zero, fall back to standard scalar_stmt
2587         cost for the rest.
2588
2589 2018-03-20  Martin Liska  <mliska@suse.cz>
2590
2591         PR ipa/84825
2592         * predict.c (rebuild_frequencies): Handle case when we have
2593         PROFILE_ABSENT, but flag_guess_branch_prob is false.
2594
2595 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2596
2597         PR target/84990
2598         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
2599         flag_section_anchors.
2600         * varasm.c (use_blocks_for_decl_p): Remove hack for
2601         dw2_force_const_mem.
2602
2603         PR target/84845
2604         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
2605         to ...
2606         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
2607         be created, use lowpart_subreg of operands[0] rather than operands[0]
2608         itself.
2609         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
2610         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
2611         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
2612         and n constraint instead of aarch64_shift_imm_di and Usd.
2613         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
2614         (*aarch64_<optab>_reg_minus<mode>3): ... this.
2615
2616 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
2617
2618         PR target/82989
2619         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
2620         to favor GPR over NEON registers.
2621         (<shift>di3_neon): Likewise.
2622
2623 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2624
2625         PR target/84952
2626         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
2627         (nvptx_process_pars): Emit bar.sync asap and alap.
2628
2629 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2630
2631         PR target/84954
2632         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
2633         seen_label if seen_label is already set.
2634
2635 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2636
2637         PR target/84945
2638         * config/i386/i386.c (fold_builtin_cpu): For features above 31
2639         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
2640         Use 1U instead of 1.  Formatting fixes.
2641
2642         PR c/84953
2643         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
2644         instead of TREE_TYPE (s1) for the return value.
2645
2646 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
2647
2648         PR tree-optimization/84946
2649         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
2650         bitsize + bitsize in poly_uint64 rather than poly_int64.
2651
2652         PR sanitizer/78651
2653         * dwarf2asm.c: Include fold-const.c.
2654         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
2655         of decl rather than decl itself.
2656
2657         PR rtl-optimization/84643
2658         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
2659
2660 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
2661
2662         PR sanitizer/78651
2663         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
2664         calling assemble_variable.
2665
2666 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
2667
2668         PR target/81647
2669         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
2670         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
2671
2672 2018-03-19  Jim Wilson  <jimw@sifive.com>
2673
2674         PR bootstrap/84856
2675         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
2676         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
2677         (riscv_first_stack_step): Likewise.
2678         (riscv_option_override): Use STACK_BOUNDARY instead of
2679         MIN_STACK_BOUNDARY.
2680         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
2681         MIN_STACK_BOUNDARY.
2682         (BIGGEST_ALIGNMENT): Set to 128.
2683         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
2684         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
2685         STACK_BOUNDARY.
2686
2687 2018-03-19  Richard Biener  <rguenther@suse.de>
2688
2689         PR tree-optimization/84933
2690         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
2691         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
2692
2693 2018-03-19  Richard Biener  <rguenther@suse.de>
2694
2695         PR tree-optimization/84859
2696         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
2697         (cond_if_else_store_replacement): Perform sinking operation on
2698         single-store BBs regardless of MAX_STORES_TO_SINK setting.
2699         Generalize what a BB with a single eligible store is.
2700
2701 2018-03-19  Richard Biener  <rguenther@suse.de>
2702
2703         PR tree-optimization/84929
2704         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
2705         chrec_is_positive against non-chrec arg.
2706
2707 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
2708
2709         PR target/84711
2710         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
2711
2712 2018-03-18  Martin Liska  <mliska@suse.cz>
2713
2714         PR rtl-optimization/84635
2715         * regrename.c (build_def_use): Use matches_mode only when
2716         matches >= 0.
2717
2718 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
2719
2720         PR tree-optimization/84913
2721         * tree-vect-loop.c (vectorizable_reduction): Don't try to
2722         vectorize chains of COND_EXPRs.
2723
2724 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2725
2726         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
2727
2728 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2729
2730         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
2731
2732 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2733
2734         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
2735
2736 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2737             Kito Cheng  <kito.cheng@gmail.com>
2738
2739         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
2740         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
2741         (nds32_adjust_reg_alloc_order): New function.
2742         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
2743
2744 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
2745
2746         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
2747         nds32_print_operand, nds32_print_operand_address): Use
2748         HOST_WIDE_INT_PRINT_DEC instead.
2749
2750 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2751
2752         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
2753
2754 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
2755
2756         PR target/84902
2757         * config/i386/i386.c (initial_ix86_tune_features,
2758         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
2759         unsigned long long.
2760         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
2761         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
2762         rather than 1u << ix86_tune.  Formatting fix.
2763         (ix86_option_override_internal): Change ix86_arch_mask from
2764         unsigned int to unsigned HOST_WIDE_INT, initialize to
2765         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
2766         (ix86_function_specific_restore): Likewise.
2767
2768 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2769
2770         PR target/84899
2771         * postreload.c (reload_combine_recognize_pattern): Perform
2772         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
2773         truncate_int_for_mode the result for the destination's mode.
2774
2775         PR c/84909
2776         * hsa-gen.c (mem_type_for_type): Fix comment typo.
2777         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
2778         Likewise.
2779         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
2780         Likewise.
2781
2782 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
2783
2784         PR target/84876
2785         * lra-assigns.c (lra_split_hard_reg_for): Don't use
2786         regno_allocno_class_array and sorted_pseudos.
2787         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
2788         insns where regno is used.
2789
2790 2018-03-16  Martin Liska  <mliska@suse.cz>
2791
2792         PR ipa/84833
2793         * multiple_target.c (create_dispatcher_calls): Redirect
2794         reference in the symbol table.
2795
2796 2018-03-16  Martin Liska  <mliska@suse.cz>
2797
2798         PR ipa/84722
2799         * multiple_target.c (create_dispatcher_calls): Redirect also
2800         an alias.
2801
2802 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2803
2804         PR c++/79937
2805         PR c++/82410
2806         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
2807         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
2808         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
2809
2810 2018-03-16  Julia Koval  <julia.koval@intel.com>
2811
2812         * doc/invoke.texi (Skylake Server): Add CLWB.
2813         Cannonlake): Remove CLWB.
2814
2815 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2816
2817         PR tree-optimization/84841
2818         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
2819         1 << 3.
2820         (FLOAT_ONE_CONST_TYPE): Define.
2821         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
2822         (sort_by_operand_rank): Put entries with higher constant_type last
2823         rather than first to match comments.
2824
2825 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
2826
2827         * config/nios2/nios2.md (movsi_internal): Fix thinko in
2828         split predicate.
2829
2830 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2831
2832         PR c++/79085
2833         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
2834         check and use address of target always.
2835
2836 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
2837
2838         PR target/84574
2839         * config/i386/i386.c (indirect_thunk_needed): Update comments.
2840         (indirect_thunk_bnd_needed): Likewise.
2841         (indirect_thunks_used): Likewise.
2842         (indirect_thunks_bnd_used): Likewise.
2843         (indirect_return_needed): New.
2844         (indirect_return_bnd_needed): Likewise.
2845         (output_indirect_thunk_function): Add a bool argument for
2846         function return.
2847         (output_indirect_thunk_function): Don't generate alias for
2848         function return thunk.
2849         (ix86_code_end): Call output_indirect_thunk_function to generate
2850         function return thunks.
2851         (ix86_output_function_return): Set indirect_return_bnd_needed
2852         and indirect_return_needed instead of indirect_thunk_bnd_needed
2853         and indirect_thunk_needed.
2854
2855 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
2856
2857         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
2858         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
2859         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
2860
2861 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
2862             Paul Hua <paul.hua.gm@gmail.com>
2863
2864         PR c/84852
2865         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
2866
2867 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
2868
2869         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
2870         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
2871         resp. SFmode cases.
2872
2873 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
2874
2875         PR target/84711
2876         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
2877         instead of GET_MODE_SIZE when comparing Units.
2878
2879 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
2880
2881         PR target/68256
2882         * varasm.c (hash_section): Return an unchangeble hash value
2883         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
2884         Return !aarch64_can_use_per_function_literal_pools_p ().
2885
2886 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2887
2888         PR target/84860
2889         * optabs.c (emit_conditional_move): Pass address of cmode's copy
2890         rather than address of cmode as last argument to prepare_cmp_insn.
2891
2892 2018-03-15  Julia Koval  <julia.koval@intel.com>
2893
2894         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
2895         F_AVX512VNNI, F_AVX512BITALG): New.
2896
2897 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
2898
2899         PR target/83451
2900         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
2901         insn for floating-point loads and stores.
2902
2903 2018-03-14  Carl Love  <cel@us.ibm.com>
2904
2905         * config/rs6000/rs6000-c.c: Add macro definitions for
2906         ALTIVEC_BUILTIN_VEC_PERMXOR.
2907         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
2908         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
2909         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
2910         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
2911         UNSPEC_VPERMXOR.
2912         * config/doc/extend.texi: Add prototypes for vec_permxor.
2913
2914 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
2915
2916         PR c/84852
2917         * diagnostic-show-locus.c (class layout_point): Convert m_line
2918         from int to linenum_type.
2919         (line_span::comparator): Use linenum "compare" function when
2920         comparing line numbers.
2921         (test_line_span): New function.
2922         (layout_range::contains_point): Convert param "row" from int to
2923         linenum_type.
2924         (layout_range::intersects_line_p): Likewise.
2925         (layout::will_show_line_p): Likewise.
2926         (layout::print_source_line): Likewise.
2927         (layout::should_print_annotation_line_p): Likewise.
2928         (layout::print_annotation_line): Likewise.
2929         (layout::print_leading_fixits): Likewise.
2930         (layout::annotation_line_showed_range_p): Likewise.
2931         (struct line_corrections): Likewise for field m_row.
2932         (line_corrections::line_corrections): Likewise for param "row".
2933         (layout::print_trailing_fixits): Likewise.
2934         (layout::get_state_at_point): Likewise.
2935         (layout::get_x_bound_for_row): Likewise.
2936         (layout::print_line): Likewise.
2937         (diagnostic_show_locus): Likewise for locals "last_line" and
2938         "row".
2939         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
2940         * input.c (selftest::test_linenum_comparisons): New function.
2941         (selftest::input_c_tests): Call it.
2942         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
2943         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
2944         * selftest.h (ASSERT_GT): New macro.
2945         (ASSERT_GT_AT): New macro.
2946         (ASSERT_LT): New macro.
2947         (ASSERT_LT_AT): New macro.
2948
2949 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
2950
2951         PR rtl-optimization/84780
2952         * combine.c (distribute_links): Don't make a link based on pc_rtx.
2953
2954 2018-03-14  Martin Liska  <mliska@suse.cz>
2955
2956         * tree.c (record_node_allocation_statistics): Use
2957         get_stats_node_kind.
2958         (get_stats_node_kind): New function extracted from
2959         record_node_allocation_statistics.
2960         (free_node): Use get_stats_node_kind.
2961
2962 2018-03-14  Richard Biener  <rguenther@suse.de>
2963
2964         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
2965         that the value-set of ANTIC_IN doesn't grow.
2966
2967         Revert
2968         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
2969         member.
2970         (BB_VISITED_WITH_VISITED_SUCCS): New define.
2971         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
2972
2973 2018-03-14  Julia Koval  <julia.koval@intel.com>
2974
2975         * config.gcc (icelake-client, icelake-server): New.
2976         (icelake): Remove.
2977         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
2978         (initial_ix86_arch_features): Ditto.
2979         (PTA_SKYLAKE): Add SGX.
2980         (PTA_ICELAKE): Remove.
2981         (PTA_ICELAKE_CLIENT): New.
2982         (PTA_ICELAKE_SERVER): New.
2983         (ix86_option_override_internal): Split up icelake on icelake client and
2984         icelake server.
2985         (get_builtin_code_for_version): Ditto.
2986         (fold_builtin_cpu): Ditto.
2987         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
2988         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
2989         * config/i386/i386.h (processor_type): Ditto.
2990         * doc/invoke.texi: Ditto.
2991
2992 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
2993
2994         PR sanitizer/83392
2995         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
2996         INTEGER_CST offset, add it together with bitpos / 8 and
2997         sign extend based on POINTER_SIZE.
2998
2999         PR target/84844
3000         Revert
3001         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
3002
3003         PR target/78090
3004         * config/i386/constraints.md (Yc): New register constraint.
3005         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
3006         Use Yc constraint for alternative 2 of operand 0.  Remove
3007         preferred_for_speed attribute.
3008
3009 2018-03-14  Richard Biener  <rguenther@suse.de>
3010
3011         PR tree-optimization/84830
3012         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
3013         with the old one to avoid oscillations.
3014
3015 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
3016
3017         PR target/83712
3018         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
3019         pseudos.
3020         (assign_by_spills): Return a flag of reload assignment failure.
3021         Do not process the reload assignment failures.  Do not spill other
3022         reload pseudos if they has the same reg class.  Update n if
3023         necessary.
3024         (lra_assign): Add a return arg.  Set up from the result of
3025         assign_by_spills call.
3026         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3027         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3028         usage_insns if it is not NULL.
3029         (spill_hard_reg_in_range): New function.
3030         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3031         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3032         function prototypes.
3033         (lra_assign): Change prototype.
3034         * lra.c (lra): Add code to deal with fails by splitting hard reg
3035         live ranges.
3036
3037 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
3038
3039         * config/riscv/riscv.opt (mrelax): New option.
3040         * config/riscv/riscv.c (riscv_file_start): Emit ".option
3041         "norelax" when riscv_mrelax is disabled.
3042         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
3043
3044 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
3045
3046         PR target/84743
3047         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
3048         reassociation for int modes.
3049
3050 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3051
3052         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
3053         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
3054         for big-endian.
3055         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
3056         * config/aarch64/aarch64-sve.md
3057         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
3058         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
3059         (*extend<mode><Vwide>2): Rename to...
3060         (aarch64_sve_extend<mode><Vwide>2): ...this.
3061         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
3062         renaming the old pattern to...
3063         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
3064         unsigned packs.
3065         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
3066         define_expand, renaming the old pattern to...
3067         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
3068         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
3069         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
3070         account when deciding which SVE instruction the optab should use.
3071         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
3072
3073 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3074
3075         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
3076         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
3077         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
3078         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
3079         (tlsdesc_small_<mode>): Turn a define_expand and use
3080         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
3081         (tlsdesc_small_advsimd_<mode>): ...this.
3082         (tlsdesc_small_sve_<mode>): New pattern.
3083
3084 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3085
3086         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
3087         (UNSPEC_UMUL_HIGHPART): New constants.
3088         (MUL_HIGHPART): New int iteraor.
3089         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
3090         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
3091         define_expand.
3092         (*<su>mul<mode>3_highpart): New define_insn.
3093
3094 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
3095
3096         PR lto/84805
3097         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
3098         incomplete types.
3099
3100 2018-03-13  Martin Liska  <mliska@suse.cz>
3101
3102         PR ipa/84658.
3103         * (sem_item_optimizer::sem_item_optimizer): Initialize new
3104         vector.
3105         (sem_item_optimizer::~sem_item_optimizer): Release it.
3106         (sem_item_optimizer::merge_classes): Register variable aliases.
3107         (sem_item_optimizer::fixup_pt_set): New function.
3108         (sem_item_optimizer::fixup_points_to_sets): Likewise.
3109         * ipa-icf.h: Declare new variables and functions.
3110
3111 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
3112
3113         PR middle-end/84834
3114         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
3115         integer_pow2p@2 and test integer_pow2p in condition.
3116         (A < 0 ? C : 0): Similarly for @1.
3117
3118         PR middle-end/84831
3119         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
3120         characters starting at p contain '\0' character, don't look beyond
3121         that.
3122
3123         PR target/84827
3124         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
3125         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
3126
3127         PR target/84828
3128         * reg-stack.c (change_stack): Change update_end var from int to
3129         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
3130         also call set_block_for_insn on the newly added insns and rescan.
3131
3132         PR target/84786
3133         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
3134         on the last operand.
3135
3136         PR c++/84704
3137         * tree.c (stabilize_reference_1): Return save_expr (e) for
3138         STATEMENT_LIST even if it doesn't have side-effects.
3139
3140 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
3141
3142         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
3143
3144 2018-03-12  Renlin Li  <renlin.li@arm.com>
3145
3146         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
3147         aarch64_output_scalar_simd_mov_immediate.
3148
3149 2018-03-12  Martin Sebor  <msebor@redhat.com>
3150
3151         PR tree-optimization/83456
3152         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
3153         for perfectly overlapping calls to memcpy.
3154         (gimple_fold_builtin_memory_chk): Same.
3155         (gimple_fold_builtin_strcpy): Handle no-warning.
3156         (gimple_fold_builtin_stxcpy_chk): Same.
3157         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
3158
3159 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
3160
3161         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
3162         parameter.  Use it for SFmode.
3163         (rs6000_function_arg_advance_1): Adjust.
3164         (rs6000_function_arg): Adjust.
3165         (rs6000_gimplify_va_arg): Pass false for that new parameter.
3166
3167 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
3168
3169         PR rtl-optimization/84169
3170         PR rtl-optimization/84780
3171         * combine.c (can_combine_p): Check for a 2-insn combination whether
3172         the destination register is used between the two insns, too.
3173
3174 2018-03-12  Richard Biener  <rguenther@suse.de>
3175
3176         PR tree-optimization/84803
3177         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
3178         for refs DR analysis didn't process.
3179
3180 2018-03-12  Richard Biener  <rguenther@suse.de>
3181
3182         PR tree-optimization/84777
3183         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
3184         force-vectorize loops ignore whether we are optimizing for size.
3185
3186 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
3187
3188         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
3189         (TARGET_MD_ASM_ADJUST): Define.
3190
3191 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
3192             Kito Cheng  <kito.cheng@gmail.com>
3193             Chung-Ju Wu  <jasonwucj@gmail.com>
3194
3195         * config/nds32/nds32.c (nds32_compute_stack_frame,
3196         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
3197         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
3198         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
3199         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
3200         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
3201         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
3202         * config/nds32/nds32.md (prologue, epilogue): Use macro
3203         NDS32_V3PUSH_AVAILABLE_P to do checking.
3204
3205 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
3206
3207         PR debug/58150
3208         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
3209         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
3210         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
3211         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
3212         addition of most attributes on !orig_type_die or the attribute not
3213         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
3214
3215 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3216             Chung-Ju Wu  <jasonwucj@gmail.com>
3217
3218         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
3219         __NDS32_VH__ macro.
3220         * config/nds32/nds32.opt (mvh): New option.
3221
3222 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3223             Chung-Ju Wu  <jasonwucj@gmail.com>
3224
3225         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
3226         function.
3227         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
3228         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
3229         definition.
3230
3231 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3232             Chung-Ju Wu  <jasonwucj@gmail.com>
3233
3234         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
3235         function.
3236         * config/nds32/nds32-multiple.md (strlensi): New pattern.
3237         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
3238
3239 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
3240             Kito Cheng  <kito.cheng@gmail.com>
3241             Chung-Ju Wu  <jasonwucj@gmail.com>
3242
3243         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
3244         UNSPEC_FFMISM and UNSPEC_FLMISM.
3245         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
3246         for ffb, ffmism and flmism.
3247         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
3248         (unspec_ffmism): Ditto.
3249         (unspec_flmism): Ditto.
3250         (nds32_expand_builtin_impl): Check if string extension is available.
3251         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
3252         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
3253
3254 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
3255
3256         Reverting patch:
3257         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
3258
3259         PR target/83712
3260         * lra-assigns.c (assign_by_spills): Return a flag of reload
3261         assignment failure.  Do not process the reload assignment
3262         failures.  Do not spill other reload pseudos if they has the same
3263         reg class.
3264         (lra_assign): Add a return arg.  Set up from the result of
3265         assign_by_spills call.
3266         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3267         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3268         usage_insns if it is not NULL.
3269         (spill_hard_reg_in_range): New function.
3270         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3271         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3272         function prototypes.
3273         (lra_assign): Change prototype.
3274         * lra.c (lra): Add code to deal with fails by splitting hard reg
3275         live ranges.
3276
3277 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
3278
3279         PR target/84807
3280         * config/i386/i386.opt: Replace Enforcment with Enforcement.
3281
3282 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
3283
3284         PR debug/84620
3285         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
3286         (dw_val_node): Add val_symbolic_view.
3287         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
3288         (symview_upper_bound): New.
3289         (new_line_info_table): Initialize symviews_since_reset.
3290         (dwarf2out_source_line): Count symviews_since_reset and set
3291         symview_upper_bound.
3292         (dw_val_equal_p): Handle symview.
3293         (add_AT_symview): New.
3294         (print_dw_val): Handle symview.
3295         (attr_checksum, attr_checksum_ordered): Likewise.
3296         (same_dw_val_p, size_of_die): Likewise.
3297         (value_format, output_die): Likewise.
3298         (add_high_low_attributes): Use add_AT_symview for entry_view.
3299         (dwarf2out_finish): Reset symview_upper_bound, clear
3300         zero_view_p.
3301
3302 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
3303
3304         PR target/83969
3305         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
3306         Add strict argument and use it.
3307         (rs6000_split_multireg_move): Update for new strict argument.
3308         (mem_operand_gpr): Disallow all non-offsettable addresses.
3309         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
3310
3311 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
3312
3313         PR target/84772
3314         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
3315         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
3316         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
3317
3318         PR c++/84767
3319         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
3320         decl, use remap_type if we want to use the type.
3321
3322 2018-03-09  Martin Sebor  <msebor@redhat.com>
3323
3324         PR tree-optimization/84526
3325         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
3326         Remove dead code.
3327         (builtin_access::generic_overlap): Be prepared to handle non-array
3328         base objects.
3329
3330 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
3331
3332         PR rtl-optimization/84682
3333         * lra-constraints.c (process_address_1): Check is_address flag
3334         for address constraints.
3335         (process_alt_operands): Likewise.
3336         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
3337         preprocess_constraints.
3338         * recog.h (preprocess_constraints): Add oploc parameter.
3339         Adjust callers.
3340         * recog.c (preprocess_constraints): Test address_operand for
3341         CT_ADDRESS constraints.
3342
3343 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
3344
3345         PR target/83712
3346         * lra-assigns.c (assign_by_spills): Return a flag of reload
3347         assignment failure.  Do not process the reload assignment
3348         failures.  Do not spill other reload pseudos if they has the same
3349         reg class.
3350         (lra_assign): Add a return arg.  Set up from the result of
3351         assign_by_spills call.
3352         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3353         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3354         usage_insns if it is not NULL.
3355         (spill_hard_reg_in_range): New function.
3356         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3357         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3358         function prototypes.
3359         (lra_assign): Change prototype.
3360         * lra.c (lra): Add code to deal with fails by splitting hard reg
3361         live ranges.
3362
3363 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3364
3365         PR target/83193
3366         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
3367         Accept complain bool parameter.  Only emit errors if it is true.
3368         (arm_parse_cpu_option_name): Likewise.
3369         (arm_target_thumb_only): Adjust callers of the above.
3370         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
3371         prototype to take a default true bool parameter.
3372         (arm_parse_arch_option_name): Likewise.
3373
3374 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
3375             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
3376
3377         PR jit/64089
3378         PR jit/84288
3379         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
3380         * configure: Regenerate.
3381         * configure.ac ("linker --version-script option"): New.
3382         ("linker soname option"): New.
3383
3384 2018-03-09  Richard Biener  <rguenther@suse.de>
3385
3386         PR tree-optimization/84775
3387         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
3388         immediate uses of predicate stmts and mark them modified.
3389
3390         Revert
3391         PR tree-optimization/84178
3392         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3393         to caller.
3394         (version_loop_for_if_conversion): Delay update_ssa call.
3395         (tree_if_conversion): Delay update_ssa until after predicate
3396         insertion.
3397
3398 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
3399
3400         PR target/84763
3401         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
3402         when the function accesses prior frames.
3403
3404 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3405
3406         PR debug/84456
3407         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
3408         gen_llsym, otherwise call maybe_gen_llsym.
3409
3410         PR inline-asm/84742
3411         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
3412         has ',' character inside of it.
3413
3414 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3415
3416         PR target/84748
3417         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
3418         as clobbering CC_REGNUM.
3419
3420 2018-03-08  Richard Biener  <rguenther@suse.de>
3421
3422         PR middle-end/84552
3423         * tree-scalar-evolution.c: Include tree-into-ssa.h.
3424         (follow_copies_to_constant): Do not follow SSA names registered
3425         for update.
3426
3427 2018-03-08  Richard Biener  <rguenther@suse.de>
3428
3429         PR tree-optimization/84178
3430         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3431         to caller.
3432         (version_loop_for_if_conversion): Delay update_ssa call.
3433         (tree_if_conversion): Delay update_ssa until after predicate
3434         insertion.
3435
3436 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
3437
3438         PR tree-optimization/84178
3439         * tree-if-conv.c (release_bb_predicate): Remove the
3440         the assertion that the stmts have NULL use_ops.
3441         Discard the statements, asserting that they haven't
3442         yet been added to a BB.
3443
3444 2018-03-08  Richard Biener  <rguenther@suse.de>
3445
3446         PR tree-optimization/84746
3447         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
3448         (phi_translate): Pass in destination ANTIC_OUT set.
3449         (phi_translate_1): Likewise.  For a simplified result lookup
3450         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
3451         (phi_translate_set): Adjust.
3452         (do_pre_regular_insertion): Likewise.
3453         (do_pre_partial_partial_insertion): Likewise.
3454
3455 2018-03-08  Martin Liska  <mliska@suse.cz>
3456
3457         PR gcov-profile/84735
3458         * doc/gcov.texi: Document usage of profile files.
3459         * gcov-io.h: Document changes in the format.
3460
3461 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
3462
3463         PR debug/84404
3464         PR debug/84408
3465         * dwarf2out.c (struct dw_line_info_table): Update comments for
3466         view == -1.
3467         (FORCE_RESET_NEXT_VIEW): New.
3468         (FORCE_RESETTING_VIEW_P): New.
3469         (RESETTING_VIEW_P): Check for -1 too.
3470         (ZERO_VIEW_P): Likewise.
3471         (new_line_info_table): Force-reset next view.
3472         (dwarf2out_begin_function): Likewise.
3473         (dwarf2out_source_line): Simplify zero_view_p initialization.
3474         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
3475         view directly.  Omit view when omitting .loc at line 0.
3476
3477 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3478
3479         PR tree-optimization/84740
3480         * tree-switch-conversion.c (process_switch): Call build_constructors
3481         only if info.phi_count is non-zero.
3482
3483         PR tree-optimization/84739
3484         * tree-tailcall.c (find_tail_calls): Check call arguments against
3485         DECL_ARGUMENTS (current_function_decl) rather than
3486         DECL_ARGUMENTS (func) when checking for tail recursion.
3487
3488 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3489
3490         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
3491         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
3492         Volker Reichelt's entry and add entries for people that perform
3493         GCC fuzzy testing and report numerous bugs.
3494
3495 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
3496
3497         PR target/82411
3498         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
3499         readonly data in sdata, if that is disabled.
3500         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
3501         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
3502         -mreadonly-in-sdata option.
3503
3504 2018-03-07  Martin Sebor  <msebor@redhat.com>
3505
3506         PR tree-optimization/84468
3507         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
3508         basic block when looking for nul assignment.
3509
3510 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
3511
3512         PR target/84277
3513         * except.h (output_function_exception_table): Adjust prototype.
3514         * except.c (output_function_exception_table): Remove FNNAME parameter
3515         and add SECTION parameter.  Ouput one part of the table at a time.
3516         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
3517         the first part of the exception table and emit unwind directives.
3518         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
3519         (i386_pe_seh_cold_init): Likewise.
3520         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
3521         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
3522         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
3523         (ix86_output_call_insn): Emit a nop in one more case for SEH.
3524         * config/i386/winnt.c: Include except.h.
3525         (struct seh_frame_state): Add reg_offset, after_prologue and
3526         in_cold_section fields.
3527         (i386_pe_seh_end_prologue): Set seh->after_prologue.
3528         (i386_pe_seh_cold_init): New function.
3529         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
3530         to seh->in_cold_section.
3531         (seh_emit_push): Record the offset of the push.
3532         (seh_emit_save): Record the offet of the save.
3533         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
3534         Test seh->after_prologue to disregard the epilogue.
3535         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
3536         (i386_pe_end_cold_function): New function.
3537
3538 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3539
3540         PR fortran/84565
3541         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
3542         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
3543
3544         PR c++/84704
3545         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
3546         on tmp_var.
3547         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
3548         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
3549
3550         PR middle-end/84723
3551         * multiple_target.c: Include tree-inline.h and intl.h.
3552         (expand_target_clones): Diagnose and fail if node->definition and
3553         !tree_versionable_function_p (node->decl).
3554
3555 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
3556
3557         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
3558         sprint_ul.
3559         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
3560         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
3561         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
3562
3563 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3564
3565         PR target/84710
3566         * combine.c (try_combine): Use reg_or_subregno instead of handling
3567         just paradoxical SUBREGs and REGs.
3568
3569 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
3570
3571          * config/arc/arc.c (arc_finalize_pic): Remove function.
3572          (arc_must_save_register): We use single base PIC register, remove
3573          checks to save/restore the PIC register.
3574          (arc_expand_prologue): Likewise.
3575          * config/arc/arc-protos.h (arc_set_default_type_attributes):
3576          Remove.
3577          (arc_verify_short): Likewise.
3578          (arc_attr_type): Likewise.
3579          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
3580          (walk_stores): Likewise.
3581          (arc_address_cost): Make it static.
3582          (arc_verify_short): Likewise.
3583          (branch_dest): Likewise.
3584          (arc_attr_type): Likewise.
3585          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
3586          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
3587          (arc_final_prescan_insn): Remove inserting the nops due to
3588          hardware hazards.  It is done in reorg step.
3589          (insn_length_variant_t): Remove.
3590          (insn_length_parameters_t): Likewise.
3591          (arc_insn_length_parameters): Likewise.
3592          (arc_get_insn_variants): Likewise.
3593          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
3594
3595 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3596
3597         PR inline-asm/84683
3598         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
3599         assertion failure.
3600
3601         PR tree-optimization/84687
3602         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
3603         on new_node->decl.
3604         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
3605
3606 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3607
3608         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
3609         Rename to ppc_speculation_barrier.
3610         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
3611         __builtin_ppc_speculation_barrier.
3612
3613 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3614
3615         PR target/84700
3616         * combine.c (combine_simplify_rtx): Don't try to simplify if
3617         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
3618         are equal to x.
3619
3620 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
3621
3622         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
3623         to 32 bytes when compiling for POWER9.
3624
3625 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3626
3627         PR target/84564
3628         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
3629         regparm >= 3 with no arg reg available also for calls with
3630         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
3631
3632         PR target/84524
3633         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
3634         orig,vex.
3635         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
3636
3637 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
3638
3639         PR target/84264
3640         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
3641
3642 2018-03-05  Richard Biener  <rguenther@suse.de>
3643
3644         PR tree-optimization/84486
3645         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
3646         When inserting a __builtin_assume_aligned call set the LHS
3647         SSA name alignment info accordingly.
3648
3649 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
3650
3651         PR tree-optimization/84114
3652         * config/aarch64/aarch64.c (aarch64_reassociation_width)
3653         Avoid reassociation of FLOAT_MODE addition.
3654
3655 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
3656
3657         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
3658         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
3659         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
3660         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
3661         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
3662         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
3663         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
3664         and -mwbnoinvd.
3665         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
3666         __builtin_ia32_wbinvd): New builtins.
3667         (SPECIAL_ARGS2): New.
3668         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
3669         (SPECIAL_ARGS2): New.
3670         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
3671         (ix86_valid_target_attribute_inner_p): Ditto.
3672         (ix86_init_mmx_sse_builtins): Add special_args2.
3673         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
3674         TARGET_WBNOINVD_P): New.
3675         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
3676         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
3677         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
3678         * config/i386/immintrin.h (_wbinvd): New intrinsic.
3679         * config/i386/pconfigintrin.h: New file.
3680         * config/i386/wbnoinvdintrin.h: Ditto.
3681         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
3682         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
3683
3684 2018-03-05  Richard Biener  <rguenther@suse.de>
3685
3686         PR tree-optimization/84670
3687         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
3688         member.
3689         (BB_VISITED_WITH_VISITED_SUCCS): New define.
3690         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
3691         (compute_antic_aux): Only assert the number of values in ANTIC_IN
3692         doesn't grow if all successors (recursively) were visited at least
3693         once.
3694
3695 2018-03-05  Richard Biener  <rguenther@suse.de>
3696
3697         PR tree-optimization/84650
3698         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
3699         if executed in the loop pipeline.
3700
3701 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
3702
3703         * doc/configfiles.texi (Configuration Files): Move info about
3704         conditionalizing $target-protos.h to...
3705         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
3706         differs from $target-protos.h.
3707
3708 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
3709             Chung-Ju Wu  <jasonwucj@gmail.com>
3710
3711         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
3712         * config/nds32/nds32-multiple.md (setmemsi): Define.
3713         * config/nds32/nds32-memory-manipulation.c
3714         (nds32_gen_dup_4_byte_to_word_value): New.
3715         (emit_setmem_word_loop): New.
3716         (emit_setmem_byte_loop): New.
3717         (nds32_expand_setmem_loop): New.
3718         (nds32_expand_setmem_loop_v3m): New.
3719         (nds32_expand_setmem_unroll): New.
3720         (nds32_expand_setmem): New.
3721
3722 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3723             Chung-Ju Wu  <jasonwucj@gmail.com>
3724
3725         * config/nds32/nds32-memory-manipulation.c
3726         (nds32_emit_load_store): New.
3727         (nds32_emit_post_inc_load_store): New.
3728         (nds32_emit_mem_move): New.
3729         (nds32_emit_mem_move_block): New.
3730         (nds32_expand_movmemsi_loop_unknown_size): New.
3731         (nds32_expand_movmemsi_loop_known_size): New.
3732         (nds32_expand_movmemsi_loop): New.
3733         (nds32_expand_movmemsi_unroll): New.
3734         (nds32_expand_movmemqi): Rename ...
3735         (nds32_expand_movmemsi): ... to this.
3736         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
3737         (movmemsi): ... to this.
3738         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
3739         (nds32_expand_movmemsi): ... to this.
3740
3741 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3742             Monk Chiang  <sh.chiang04@gmail.com>
3743             Chung-Ju Wu  <jasonwucj@gmail.com>
3744
3745         * config/nds32/nds32-protos.h
3746         (nds32_expand_load_multiple): New arguments.
3747         (nds32_expand_store_multiple): Ditto.
3748         (nds32_valid_multiple_load_store): Rename ...
3749         (nds32_valid_multiple_load_store_p): ... to this.
3750         * config/nds32/nds32-memory-manipulation.c
3751         (nds32_expand_load_multiple): Refine implementation.
3752         (nds32_expand_store_multiple): Ditto.
3753         * config/nds32/nds32-multiple.md
3754         (load_multiple): Update nds32_expand_load_multiple interface.
3755         (store_multiple): Update nds32_expand_store_multiple interface.
3756         * config/nds32/nds32-predicates.c
3757         (nds32_valid_multiple_load_store): Rename ...
3758         (nds32_valid_multiple_load_store_p): ... to this and refine
3759         implementation.
3760         * config/nds32/predicates.md
3761         (nds32_load_multiple_and_update_address_operation): New predicate.
3762         (nds32_store_multiple_and_update_address_operation): New predicate.
3763
3764 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3765             Chung-Ju Wu  <jasonwucj@gmail.com>
3766
3767         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
3768         (combo): New attribute.
3769         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
3770
3771 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
3772
3773         * config/nds32/nds32.opt: Change -mcmodel= default value.
3774
3775 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
3776             Monk Chiang  <sh.chiang04@gmail.com>
3777             Chung-Ju Wu  <jasonwucj@gmail.com>
3778
3779         * config/nds32/constants.md (unspec_element): New enum.
3780         * config/nds32/constraints.md (Umw): New constraint.
3781         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
3782         * config/nds32/nds32-intrinsic.md: Likewise.
3783         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
3784         (nds32_valid_smw_lwm_base_p): New.
3785         (nds32_output_smw_single_word): New.
3786         (nds32_output_lmw_single_word): New.
3787         (nds32_expand_unaligned_load): New.
3788         (nds32_expand_unaligned_store): New.
3789         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
3790         (nds32_output_smw_single_word): Declare.
3791         (nds32_output_lmw_single_word): Declare.
3792         (nds32_expand_unaligned_load): Declare.
3793         (nds32_expand_unaligned_store): Declare.
3794         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
3795         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
3796         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
3797         NDS32_BUILTIN_UASTORE_DW.
3798         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
3799         predicate.
3800
3801 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
3802             Kito Cheng  <kito.cheng@gmail.com>
3803             Chung-Ju Wu  <jasonwucj@gmail.com>
3804
3805         * config/nds32/nds32-intrinsic.c
3806         (nds32_expand_builtin_null_ftype_reg): Delete.
3807         (nds32_expand_builtin_reg_ftype_imm): Ditto.
3808         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
3809         (nds32_read_argument): New.
3810         (nds32_legitimize_target): Ditto.
3811         (nds32_legitimize_argument): Ditto.
3812         (nds32_check_constant_argument): Ditto.
3813         (nds32_expand_unop_builtin): Ditto.
3814         (nds32_expand_unopimm_builtin): Ditto.
3815         (nds32_expand_binop_builtin): Ditto.
3816         (nds32_builtin_decl_impl): Ditto.
3817         (builtin_description): Ditto.
3818         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
3819         (nds32_init_builtins_impl): Ditto.
3820         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
3821         (nds32_builtin_decl): New.
3822         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
3823         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
3824
3825 2018-03-02  Jeff Law  <law@redhat.com>
3826
3827         * reorg.c (stop_search_p): Handle DEBUG_INSN.
3828         (redundant_insn, fill_simple_delay_slots): Likewise.
3829         (fill_slots_from_thread): Likewise.
3830         * resource.c (mark_referenced_resources): Likewise.
3831         (mark_set_resources, find_dead_or_set_registers): Likewise.
3832
3833 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3834
3835         * substring-locations.h (format_warning_va): Formatting fix for
3836         ATTRIBUTE_GCC_DIAG.
3837         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
3838         argument.
3839         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
3840         * substring-locations.c: Include intl.h.
3841         (format_warning_va): Turned into small wrapper around
3842         format_warning_n_va, renamed to ...
3843         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
3844         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
3845         use ngettext.
3846         (format_warning_at_substring_n): New function.
3847         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
3848         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
3849         format_warning_at_substring with just a shorter name instead of
3850         const function pointer.
3851         (fmtwarn_n): New function.
3852         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
3853         appropriate, get rid of all the fmtstr temporaries, move conditionals
3854         with G_() wrapped string literals directly into fmtwarn arguments,
3855         cast dir.len to (int), formatting fixes.
3856
3857 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
3858
3859         * doc/invoke.texi: Remove "Cilk Plus" references.
3860
3861 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3862             Richard Biener  <rguenther@suse.de>
3863
3864         PR ipa/84628
3865         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
3866         for error or warning attributes if CALL_FROM_THUNK_P is set.
3867         Formatting fixes.
3868
3869 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3870
3871         PR target/56540
3872         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
3873         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
3874
3875         PR target/56540
3876         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
3877         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
3878
3879         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
3880         instead of -1U in last predictors element's probability member.
3881
3882 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
3883
3884         PR ipa/83983
3885         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
3886         arguments if they are comparable.
3887
3888 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3889
3890         PR tree-optimization/84634
3891         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
3892         masks and masked_loop_p with a single loop_masks, making sure it's
3893         null for bb vectorization.
3894
3895 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3896
3897         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
3898         (vect_analyze_data_ref_access): Use loop->safe_len rather than
3899         loop->force_vectorize to check whether there is no alias.
3900
3901 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3902
3903         PR target/84614
3904         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
3905         prototypes.
3906         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
3907         comments.
3908         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
3909         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
3910         instead of a loop around prev_real_insn.
3911         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
3912         prev_real_insn.
3913
3914         PR inline-asm/84625
3915         * config/i386/i386.c (ix86_print_operand): Use conditional
3916         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
3917         zero vector.
3918
3919 2018-03-02  Richard Biener  <rguenther@suse.de>
3920
3921         PR tree-optimization/84427
3922         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
3923         (bitmap_set_subtract_values): Rewrite to handle multiple
3924         exprs per value.
3925         (clean): Likewise.
3926         (prune_clobbered_mems): Likewise.
3927         (phi_translate): Take edge instead of pred/phiblock.
3928         (phi_translate_1): Likewise.
3929         (phi_translate_set): Likewise.  Insert all translated
3930         exprs for a value into the set, keeping possibly multiple
3931         expressions per value.
3932         (compute_antic_aux): Adjust for phi_translate changes.
3933         When intersecting union the expressions and prune those
3934         not in the final value set, keeping possibly multiple
3935         expressions per value.  Do not use value-insertion
3936         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
3937         all expressions.  Add verification that the value-sets
3938         only shrink during iteration.
3939         (compute_partial_antic_aux): Adjust for the phi_translate changes.
3940         (do_pre_regular_insertion): Likewise.
3941         (do_pre_partial_partial_insertion): Likewise.
3942
3943 2018-03-02  Richard Biener  <rguenther@suse.de>
3944
3945         PR target/82005
3946         * config/darwin.c (saved_debug_info_level): New static global.
3947         (darwin_asm_lto_start): Disable debug info generation for LTO out.
3948         (darwin_asm_lto_end): Restore debug info generation settings.
3949
3950 2018-03-01  Martin Liska  <mliska@suse.cz>
3951
3952         PR sanitizer/82484
3953         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
3954         volatile arguments.
3955
3956 2018-03-01  Richard Biener  <rguenther@suse.de>
3957
3958         PR debug/84645
3959         * dwarf2out.c (gen_variable_die): Properly handle late VLA
3960         type annotation with LTO when debug was disabled at compile-time.
3961
3962 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
3963
3964         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
3965         XINT with INTVAL.
3966         (mips_final_postscan_insn): Likewise.
3967
3968 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
3969
3970         PR rtl-optimization/84528
3971         * alias.c (init_alias_target): Add commentary.
3972         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
3973         a unique base value if the frame pointer is not eliminated
3974         to the stack pointer.
3975
3976 2018-03-01  Tom de Vries  <tom@codesourcery.com>
3977
3978         PR rtl-optimization/83327
3979         * lra-int.h (hard_regs_spilled_into): Declare.
3980         * lra.c (hard_regs_spilled_into): Define.
3981         (init_reg_info): Init hard_regs_spilled_into.
3982         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
3983         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
3984         (process_bb_lives): Handle hard_regs_spilled_into.
3985         (lra_create_live_ranges_1): Before doing liveness propagation, clear
3986         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
3987
3988 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
3989
3990         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
3991         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
3992         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
3993         * config/rs6000/aix72.h: New file.
3994
3995 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
3996
3997         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
3998         instead of warning_at with conditional singular and plural messages
3999         where possible.
4000
4001         PR target/52991
4002         * stor-layout.c (update_alignment_for_field): For
4003         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
4004         && !DECL_PACKED (field), do the alignment update, just use
4005         only desired_align instead of MAX (type_align, desired_align)
4006         as the alignment.
4007         (place_field): Don't do known_align < desired_align handling
4008         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
4009         is non-NULL, instead do it after rli->prev_field handling and
4010         only if not within a bitfield word.  For DECL_PACKED (field)
4011         use type_align of BITS_PER_UNIT.
4012
4013 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
4014
4015         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
4016         superfluous parentheses and trailing spaces.
4017
4018 2018-02-28  Richard Biener  <rguenther@suse.de>
4019
4020         PR tree-optimization/84584
4021         * graphite-scop-detection.c (scop_detection::add_scop): Discard
4022         SCoPs with fake exit edge.
4023
4024 2018-02-28  Martin Liska  <mliska@suse.cz>
4025
4026         PR testsuite/84597
4027         * timevar.c (timer::print): Fix format to properly print 100%
4028         values.
4029
4030 2018-02-28  Richard Biener  <rguenther@suse.de>
4031
4032         PR middle-end/84607
4033         * genmatch.c (capture_info::walk_match): Do not mark
4034         captured expressions without operands as expr_p given
4035         they act more like predicates and should be subject to
4036         "lost tail" side-effect preserving.
4037
4038 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
4039
4040         PR rtl-optimization/81611
4041         * auto-inc-dec.c (attempt_change): Move dead note from
4042         mem_insn if it's the next use of regno
4043         (find_address): Take address use of reg holding
4044         non-incremented value.  Add parm to limit search to the named
4045         reg only.
4046         (merge_in_block): Attempt to use a mem insn that is the next
4047         use of the original regno.
4048
4049 2018-02-27  Martin Sebor  <msebor@redhat.com>
4050
4051         PR c++/83871
4052         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
4053         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
4054
4055 2018-02-27  Martin Sebor  <msebor@redhat.com>
4056
4057         PR translation/84207
4058         * diagnostic-core.h (warning_n, error_n, inform_n): Change
4059         n argument to unsigned HOST_WIDE_INT.
4060         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
4061         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
4062         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
4063         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
4064
4065 2018-02-27  Richard Biener  <rguenther@suse.de>
4066
4067         PR tree-optimization/84512
4068         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
4069         Do not use the estimate returned from record_stmt_cost for
4070         the scalar iteration cost but sum properly using add_stmt_cost.
4071
4072 2018-02-27  Richard Biener  <rguenther@suse.de>
4073
4074         PR tree-optimization/84466
4075         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4076         Adjust last change to less strictly validate use operands.
4077
4078 2018-02-27  Martin Liska  <mliska@suse.cz>
4079
4080         PR gcov-profile/84548
4081         * gcov.c (process_file): Allow partial overlap and consider it
4082         also as group functions.
4083         (output_lines): Properly calculate range of lines for a group.
4084
4085 2018-02-27  Martin Liska  <mliska@suse.cz>
4086
4087         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
4088         'ggc' suffixes.  Change first column width.
4089         (timer::print): Fix formatting of the column.
4090
4091 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
4092
4093         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
4094         preserve inline entry blocks for the sake of debug inline
4095         entry point markers alone.
4096         (remove_unused_locals): Suggest in comments a better place to
4097         force the preservation of inline entry blocks that are
4098         otherwise unused, but do not preserve them.
4099
4100 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4101
4102         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
4103
4104 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4105
4106         PR target/84039
4107         * config/i386/constraints.md (Bs): Replace
4108         ix86_indirect_branch_register with
4109         TARGET_INDIRECT_BRANCH_REGISTER.
4110         (Bw): Likewise.
4111         * config/i386/i386.md (indirect_jump): Likewise.
4112         (tablejump): Likewise.
4113         (*sibcall_memory): Likewise.
4114         (*sibcall_value_memory): Likewise.
4115         Peepholes of indirect call and jump via memory: Likewise.
4116         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
4117         (*sibcall_value_GOT_32): Likewise.
4118         * config/i386/predicates.md (indirect_branch_operand): Likewise.
4119         (GOT_memory_operand): Likewise.
4120         (call_insn_operand): Likewise.
4121         (sibcall_insn_operand): Likewise.
4122         (GOT32_symbol_operand): Likewise.
4123         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
4124
4125 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
4126
4127         PR rtl-optimization/83496
4128         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
4129         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
4130         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
4131         redundant insn, if any.
4132         (relax_delay_slots): Likewise.
4133         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
4134
4135 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
4136
4137         PR tree-optimization/83965
4138         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
4139         that grouped statements are part of a reduction chain.  Return
4140         true if the statement is not marked as a reduction itself but
4141         is part of a group.
4142         (vect_recog_dot_prod_pattern): Don't check whether the statement
4143         is part of a group here.
4144         (vect_recog_sad_pattern): Likewise.
4145         (vect_recog_widen_sum_pattern): Likewise.
4146
4147 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
4148
4149         PR debug/84545
4150         * final.c (rest_of_clean_state): Also look for calls inside sequences.
4151
4152 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4153
4154         PR target/84530
4155         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
4156         the bool argument.
4157         (ix86_output_indirect_function_return): New prototype.
4158         (ix86_split_simple_return_pop_internal): Likewise.
4159         * config/i386/i386.c (indirect_return_via_cx): New.
4160         (indirect_return_via_cx_bnd): Likewise.
4161         (indirect_thunk_name): Handle return va CX_REG.
4162         (output_indirect_thunk_function): Create alias for
4163         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
4164         (ix86_output_indirect_jmp): Remove the bool argument.
4165         (ix86_output_indirect_function_return): New function.
4166         (ix86_split_simple_return_pop_internal): Likewise.
4167         * config/i386/i386.md (*indirect_jump): Don't pass false
4168         to ix86_output_indirect_jmp.
4169         (*tablejump_1): Likewise.
4170         (simple_return_pop_internal): Change it to define_insn_and_split.
4171         Call ix86_split_simple_return_pop_internal to split it for
4172         -mfunction-return=.
4173         (simple_return_indirect_internal): Call
4174         ix86_output_indirect_function_return instead of
4175         ix86_output_indirect_jmp.
4176
4177 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
4178
4179         PR bootstrap/84405
4180         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
4181         memset and value initialization afterwards.
4182
4183 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
4184
4185         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
4186
4187 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4188
4189         PR target/84521
4190         * common/config/aarch64/aarch64-common.c
4191         (aarch_option_optimization_table[]): Switch
4192         off fomit-frame-pointer
4193
4194 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
4195             Chung-Ju Wu  <jasonwucj@gmail.com>
4196
4197         * config/nds32/nds32-multiple.md (load_multiple): Disallow
4198         volatile memory.
4199         (store_multiple): Ditto.
4200
4201 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
4202
4203         * config.gcc: Add --with-cpu support for nds32 target.
4204         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
4205         * config/nds32/nds32.opt: Add -mcpu= option.
4206
4207 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
4208
4209         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
4210         isel=yes): Warn for these deprecated options.
4211
4212 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
4213
4214         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
4215         ISA_2_5_MASKS_EMBEDDED.
4216
4217 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
4218
4219         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
4220         p->max as pointers rather than using iterative_hash_expr.
4221
4222 2018-02-23  Carl Love  <cel@us.ibm.com>
4223
4224         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
4225         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
4226         BU_P8V_OVERLOAD_2.
4227         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
4228         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
4229         P8V_BUILTIN_VEC_VUNSIGNED2.
4230
4231 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
4232
4233         PR target/81572
4234         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
4235         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
4236         LRA_UNKNOWN_ALT.
4237         * lra-constraints.c (curr_insn_transform): Set up
4238         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
4239         LRA_UNKNOWN_ALT.
4240         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
4241         * lra-eliminations.c (spill_pseudos): Ditto.
4242         (process_insn_for_elimination): Ditto.
4243         * lra-lives.c (reg_early_clobber_p): Use the new macros.
4244         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
4245         LRA_NON_CLOBBERED_ALT.
4246
4247 2018-02-22  Martin Sebor  <msebor@redhat.com>
4248
4249         PR tree-optimization/84480
4250         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
4251         to maybe_diag_stxncpy_trunc.  Call it.
4252         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
4253         from gimple_fold_builtin_strcpy.  Print inlining stack.
4254         (handle_builtin_stxncpy): Print inlining stack.
4255         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
4256
4257 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
4258
4259         PR target/84176
4260         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
4261         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
4262         and -fcheck-pointer-bounds are used together.
4263         (indirect_thunk_prefix): New enum.
4264         (indirect_thunk_need_prefix): New function.
4265         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
4266         "_nt" instead of "_bnd" for NOTRACK prefix.
4267         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
4268         (output_indirect_thunk_function): Likewise.
4269         (): Likewise.
4270         (ix86_code_end): Update output_indirect_thunk_function calls.
4271         (ix86_output_indirect_branch_via_reg): Replace
4272         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
4273         (ix86_output_indirect_branch_via_push): Likewise.
4274         (ix86_output_function_return): Likewise.
4275         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
4276         incompatible with -fcf-protection=branch and
4277         -fcheck-pointer-bounds.
4278
4279 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
4280
4281         PR target/83335
4282         * config/aarch64/aarch64.c (aarch64_print_address_internal):
4283         Change gcc_assert call to output_operand_lossage.
4284
4285 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
4286
4287         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
4288
4289 2018-02-22  DJ Delorie  <dj@redhat.com>
4290             Sebastian Perta  <sebastian.perta@renesas.com>
4291             Oleg Endo  <olegendo@gcc.gnu.org>
4292
4293         * config/rx/rx.c (rx_rtx_costs): New function.
4294         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
4295
4296 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
4297
4298         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
4299
4300 2018-02-22  Martin Liska  <mliska@suse.cz>
4301
4302         PR driver/83193
4303         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
4304         Add "native" as a possible value.
4305
4306 2018-02-22  Martin Liska  <mliska@suse.cz>
4307
4308         PR driver/83193
4309         * config/i386/i386.c (ix86_option_override_internal):
4310         Add "native" as a possible value for -march and -mtune.
4311
4312 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
4313
4314         PR target/84502
4315         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
4316         to all type variants.
4317
4318         PR tree-optimization/84503
4319         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
4320         width as info->bitpos + info->bitsize - start.
4321         (merged_store_group::merge_overlapping): Simplify width computation.
4322         (check_no_overlap): New function.
4323         (imm_store_chain_info::try_coalesce_bswap): Compute expected
4324         start + width and last_order of the group, fail if check_no_overlap
4325         fails.
4326         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
4327         to group if check_no_overlap fails.
4328
4329 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
4330
4331         * config/rs6000/altivec.md: Delete contraint arguments to
4332         define_expand, define_split, and define_peephole2, and in
4333         define_insn_and_split if always unused.
4334         * config/rs6000/darwin.md: Ditto.
4335         * config/rs6000/dfp.md: Ditto.
4336         * config/rs6000/rs6000.md: Ditto.
4337         * config/rs6000/sync.md: Ditto.
4338         * config/rs6000/vector.md: Ditto.
4339         * config/rs6000/vsx.md: Ditto.
4340
4341 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
4342
4343         * config/rs6000/altivec.md: Write output control strings as braced
4344         blocks instead of double-quoted strings.
4345         * config/rs6000/darwin.md: Ditto.
4346         * config/rs6000/rs6000.md: Ditto.
4347         * config/rs6000/vector.md: Ditto.
4348         * config/rs6000/vsx.md: Ditto.
4349
4350 2018-02-21  Jason Merrill  <jason@redhat.com>
4351
4352         PR c++/84314 - ICE with templates and fastcall attribute.
4353         * attribs.c (build_type_attribute_qual_variant): Remove assert.
4354
4355 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
4356
4357         * ipa-cp.c (determine_versionability): Fix comment typos.
4358
4359 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
4360
4361         PR c/84229
4362         * ipa-cp.c (determine_versionability): Do not version functions caling
4363         va_arg_pack.
4364
4365 2018-02-21  Martin Liska  <mliska@suse.cz>
4366
4367         PR driver/83193
4368         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
4369         Add "native" as a possible value.
4370         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
4371         the macro when native cpu detection is available.
4372
4373 2018-02-21  Martin Liska  <mliska@suse.cz>
4374
4375         PR driver/83193
4376         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
4377         Add "native" as a possible value.
4378         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
4379         when native cpu detection is available.
4380
4381 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
4382             Martin Sebor  <msebor@redhat.com>
4383
4384         PR tree-optimization/84478
4385         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
4386         false.
4387         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
4388         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
4389         support which is conservatively correct, for 2 only stay conservative
4390         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
4391         argument to the 2 argument get_range_strlen, adjust 6 arg
4392         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
4393         false.
4394         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
4395         (gimple_fold_builtin_strlen): Pass true as last argument to
4396         get_range_strlen.
4397
4398 2018-02-20  Martin Sebor  <msebor@redhat.com>
4399
4400         PR middle-end/84095
4401         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
4402         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
4403         (builtin_memref::builtin_memref): Factor out parts into
4404         set_base_and_offset and call it.
4405
4406 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4407
4408         PR middle-end/84406
4409         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
4410         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
4411         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
4412         search at the associated MODE_INT.
4413
4414 2018-02-20  Jeff Law  <law@redhat.com>
4415
4416         PR middle-end/82123
4417         PR tree-optimization/81592
4418         PR middle-end/79257
4419         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
4420         for range data rather than using global data.
4421         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
4422         range data rather than using global data.
4423         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
4424         pass it to children as needed.
4425         (struct directive::fmtresult): Similarly.
4426         (struct directive::set_width): Similarly.
4427         (struct directive::set_precision): Similarly.
4428         (format_integer, format_directive, parse_directive): Similarly.
4429         (format_none): Accept unnamed vr_values parameter.
4430         (format_percent, format_floating, format_character): Similarly.
4431         (format_string, format_plain): Similarly.
4432         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
4433         the EVRP range analyzer for range data rather than using global data.
4434         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
4435         gimple-ssa-evrp-analyze.h
4436         (class sprintf_dom_walker): Add after_dom_children member function.
4437         Add evrp_range_analyzer member.
4438         (sprintf_dom_walker::before_dom_children): Call into the EVRP
4439         range analyzer as needed.
4440         (sprintf_dom_walker::after_dom_children): New member function.
4441         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
4442         if not optimizing.
4443         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
4444         (evrp_range_analyzer::pop_to_marker): Likewise.
4445
4446 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4447
4448         PR tree-optimization/84419
4449         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
4450         with the required type if its current type is compatible but
4451         different.
4452
4453 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
4454
4455         PR middle-end/82004
4456         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
4457         after vectorization.
4458
4459 2018-02-20  Martin Liska  <mliska@suse.cz>
4460
4461         PR driver/83193
4462         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
4463         possible values if we don't have a hint.
4464
4465 2018-02-20  Martin Liska  <mliska@suse.cz>
4466
4467         PR c/84310
4468         PR target/79747
4469         * final.c (shorten_branches): Build align_tab array with one
4470         more element.
4471         * opts.c (finish_options): Add alignment option limit check.
4472         (MAX_CODE_ALIGN): Likewise.
4473         (MAX_CODE_ALIGN_VALUE): Likewise.
4474         * doc/invoke.texi: Document maximum allowed option value for
4475         all -falign-* options.
4476
4477 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
4478
4479         PR target/84146
4480         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
4481         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
4482         * var-tracking.c (emit_note_insn_var_location): Remove all references
4483         to NOTE_INSN_CALL_ARG_LOCATION.
4484         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
4485         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
4486         Use copy_rtx_if_shared.
4487         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
4488         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
4489         (dwarf2out_var_location): Remove handling of
4490         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
4491         on call_insn.
4492         * final.c (final_scan_insn): Remove all references to
4493         NOTE_INSN_CALL_ARG_LOCATION.
4494         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
4495         before dumping final insns.
4496         * except.c (emit_note_eh_region_end): Remove all references to
4497         NOTE_INSN_CALL_ARG_LOCATION.
4498         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
4499         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
4500         * config/arc/arc.c (hwloop_optimize): Likewise.
4501         * config/arm/arm.c (create_fix_barrier): Likewise.
4502         * config/s390/s390.c (s390_chunkify_start): Likewise.
4503         * config/sh/sh.c (find_barrier): Likewise.
4504         * config/i386/i386.c (rest_of_insert_endbranch,
4505         ix86_seh_fixup_eh_fallthru): Likewise.
4506         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
4507         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
4508         * config/frv/frv.c (frv_function_prologue): Likewise.
4509         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
4510         reg note.
4511         (note_outside_basic_block_p): Remove all references to
4512         NOTE_INSN_CALL_ARG_LOCATION.
4513         * gengtype.c (adjust_field_rtx_def): Likewise.
4514         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
4515         Likewise.
4516         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
4517         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
4518
4519         PR c++/84444
4520         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
4521         is ADDR_EXPR.
4522
4523         PR tree-optimization/84452
4524         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
4525         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
4526         is NULL.
4527
4528 2018-02-19  Martin Liska  <mliska@suse.cz>
4529
4530         PR sanitizer/82183
4531         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
4532
4533 2018-02-19  Martin Liska  <mliska@suse.cz>
4534             Richard Sandiford  <richard.sandiford@linaro.org>
4535
4536         PR tree-optimization/82491
4537         * gimple-fold.c (get_base_constructor): Make earlier bail out
4538         to prevent ubsan.
4539
4540 2018-02-19  Carl Love  <cel@us.ibm.com>
4541
4542         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
4543         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
4544         BU_P8V_OVERLOAD_1.
4545         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
4546         P8V_BUILTIN_VEC_NEG.
4547
4548 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
4549
4550         * config/rl78/rl78.md (movdf): New define expand.
4551
4552 2018-02-19  Martin Liska  <mliska@suse.cz>
4553
4554         PR other/80589
4555         * doc/invoke.texi: Fix typo.
4556         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
4557
4558 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
4559
4560         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
4561         handle rs6000_single_float and rs6000_double_float specially for
4562         e500 family CPUs.
4563
4564 2018-02-16  Jeff Law  <law@redhat.com>
4565
4566         * config/rx/rx.c (add_pop_cfi_notes): New function.;
4567         (pop_regs): Use it.
4568
4569 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4570
4571         PR ipa/84425
4572         * ipa-inline.c (inline_small_functions): Fix a typo.
4573
4574 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4575
4576         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
4577
4578 2018-02-16  Carl Love  <cel@us.ibm.com>
4579
4580         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
4581         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
4582         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
4583         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
4584         expansion to P8V_BUILTIN_VEC_FLOAT2.
4585
4586 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
4587
4588         PR rtl-optimization/70023
4589         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
4590         src_regno into account.
4591
4592 2018-02-16  Carl Love  <cel@us.ibm.com>
4593
4594         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
4595         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
4596         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
4597         * config/rs6000/rs6000.c: Remove case statements for
4598         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
4599         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
4600         and P9V_BUILTIN_VEC_VINSERT4B.
4601         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
4602         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
4603         * config/rs6000/vsx.md:
4604         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
4605         vec_insert4b.
4606
4607 2018-02-16  Carl Love  <cel@us.ibm.com>
4608
4609         * config/rs6000/altivec.h: Add builtin names vec_extract4b
4610         vec_insert4b.
4611         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
4612         definitions.
4613         * config/rs6000/rs6000-c.c: Add the definitions for
4614         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
4615         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
4616         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
4617         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
4618         definition for insert4b and define insn *insert3b_internal.
4619         * doc/extend.texi: Add documentation for vec_extract4b.
4620
4621 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4622
4623         * doc/extend.texi (Backwards Compatibility): Mention friend
4624         injection.  Note for-scope is deprecated.
4625         * doc/invoke.texi (-ffriend-injection): Deprecate.
4626
4627 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
4628
4629         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
4630         that moved to I2, also allow destinations that are a paradoxical
4631         subreg (instead of a normal reg).
4632
4633 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
4634
4635         PR target/83831
4636         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
4637         to QImode.
4638
4639 2018-02-16  Richard Biener  <rguenther@suse.de>
4640
4641         PR tree-optimization/84037
4642         PR tree-optimization/84016
4643         PR target/82862
4644         * config/i386/i386.c (ix86_builtin_vectorization_cost):
4645         Adjust vec_construct for the fact we need additional higher latency
4646         128bit inserts for AVX256 and AVX512 vector builds.
4647         (ix86_add_stmt_cost): Scale vector construction cost for
4648         elementwise loads.
4649
4650 2018-02-16  Richard Biener  <rguenther@suse.de>
4651
4652         PR tree-optimization/84417
4653         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
4654         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
4655         (non_rewritable_lvalue_p): Likewise, use poly-ints.
4656
4657 2018-02-16  Martin Liska  <mliska@suse.cz>
4658
4659         PR sanitizer/84307
4660         * internal-fn.def (ASAN_CHECK): Set proper flags.
4661         (ASAN_MARK): Likewise.
4662
4663 2018-02-16  Julia Koval  <julia.koval@intel.com>
4664
4665         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
4666         from PTA_CANNONLAKE.
4667
4668 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4669
4670         PR target/84272
4671         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
4672         Use ++iter rather than iter++ for std::list iterators.
4673         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
4674         defer deleting them until all nodes in the forest are processed.  Do
4675         free even leaf nodes.  Change to_process into auto_vec.
4676
4677         PR bootstrap/84405
4678         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
4679         * vec.h (vec_default_construct): Use memset instead of placement new
4680         if BROKEN_VALUE_INITIALIZATION is defined.
4681         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
4682         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
4683         is defined.
4684
4685         PR rtl-optimization/83723
4686         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
4687         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
4688         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
4689         recursive calls.
4690         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
4691         callers.
4692         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
4693
4694 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
4695
4696         PR rtl-optimization/81443
4697         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
4698         from inner REGs to paradoxical SUBREGs.
4699
4700 2018-02-16  Richard Biener  <rguenther@suse.de>
4701
4702         PR tree-optimization/84399
4703         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4704         For operands we can analyze at their definition make sure we can
4705         analyze them at each use as well.
4706
4707 2018-02-16  Richard Biener  <rguenther@suse.de>
4708
4709         PR tree-optimization/84190
4710         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
4711         volatile accesses if the decl isn't volatile.
4712
4713 2018-02-15  Jason Merrill  <jason@redhat.com>
4714
4715         PR c++/84314 - ICE with templates and fastcall attribute.
4716         * attribs.c (build_type_attribute_qual_variant): Don't clobber
4717         TYPE_CANONICAL on an existing type.
4718
4719 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
4720
4721         PR tree-optimization/84383
4722         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
4723         dstoff nor call operand_equal_p if dstbase is NULL.
4724
4725         PR tree-optimization/84334
4726         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
4727         also a CONSTANT_CLASS_P, punt.
4728
4729 2018-02-14  Jim Wilson  <jimw@sifive.com>
4730
4731         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
4732         first SMALL_OPERAND check.  New local min_second_step.  Move assert
4733         to where locals are set.  Add TARGET_RVC support.
4734         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
4735
4736 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
4737
4738         * doc/invoke.texi: Correct -Wformat-overflow code sample.
4739
4740 2018-02-14  Martin Sebor  <msebor@redhat.com>
4741
4742         PR tree-optimization/83698
4743         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
4744         arrays constrain the offset range to their bounds.
4745         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
4746         (builtin_access::overlap): Avoid setting the size of overlap if it's
4747         already been set.
4748         (maybe_diag_overlap): Also consider arrays when deciding what values
4749         of offsets to include in diagnostics.
4750
4751 2018-02-14  Martin Sebor  <msebor@redhat.com>
4752
4753         PR c/84108
4754         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
4755         that correspond to the kind of a declaration.
4756
4757 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
4758
4759         PR target/83984
4760         * config/pa/pa.md: Load address of PIC label using the linkage table
4761         if the label is nonlocal.
4762
4763 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4764
4765         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
4766         warning message if user requests -maltivec=be.
4767         * doc/invoke.texi: Document deprecation of -maltivec=be.
4768
4769 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
4770
4771         PR target/84220
4772         * config/rs6000/rs6000-c.c: Update definitions for
4773         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
4774         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
4775
4776 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
4777
4778         PR target/84239
4779         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
4780         add _get_ssp intrinsics. Remove argument from
4781         __builtin_ia32_rdssp[d|q].
4782         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
4783         * config/i386/i386-builtin.def: Remove argument from
4784         __builtin_ia32_rdssp[d|q].
4785         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
4786         ix86_expand_special_args_builtin for _rdssp[d|q].
4787         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
4788         Clear register before usage.
4789         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
4790         Add documentation for new _get_ssp and _inc_ssp intrinsics.
4791
4792 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
4793
4794         PR tree-optimization/84357
4795         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
4796         operand 1 of an ARRAY_REF too.
4797
4798 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
4799
4800         PR target/83831
4801         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
4802         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4803         declarations.
4804         (set_of_reg): New struct.
4805         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
4806         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
4807         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4808         functions.
4809         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
4810         Split into bitclr, bitset, bitinvert patterns if appropriate.
4811         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
4812         use rx_fuse_in_memory_bitop.
4813         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
4814         to named insn, correct maximum insn length.
4815
4816 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
4817
4818         PR target/79242
4819         * machmode.def: Define a complex mode for PARTIAL_INT.
4820         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
4821         MODE_PARTIAL_INT.
4822         * doc/rtl.texi: Document CSPImode.
4823         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
4824         handling.
4825         (msp430_hard_regno_nregs_with_padding): Likewise.
4826
4827 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
4828
4829         PR target/84279
4830         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
4831
4832 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
4833
4834         PR rtl-optimization/84169
4835         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
4836         we generated a parallel as new i3 and we split that to new i2 and i3
4837         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
4838         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
4839         those to i2, not i1.  Partially rewrite this scan code.
4840
4841 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4842
4843         PR c/82210
4844         * stor-layout.c (place_field): For variable length fields, adjust
4845         offset_align afterwards not just based on the field's alignment,
4846         but also on the size.
4847
4848         PR middle-end/84309
4849         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
4850         of exps and logs in the use_exp2 case.
4851
4852 2018-02-13  Jeff Law  <law@redhat.com>
4853
4854         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
4855         entry for "vector".
4856
4857         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
4858         ARGS as unused.
4859
4860 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
4861
4862         PR debug/84342
4863         PR debug/84319
4864         * common.opt (gas-loc-support, gas-locview-support): New.
4865         (ginline-points, ginternal-reset-location-views): New.
4866         * doc/invoke.texi: Document them.  Use @itemx where intended.
4867         (gvariable-location-views): Adjust.
4868         * target.def (reset_location_view): New.
4869         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
4870         (TARGET_RESET_LOCATION_VIEW): New.
4871         * doc/tm.texi: Rebuilt.
4872         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
4873         (dwarf2out_default_as_locview_support): New.
4874         (output_asm_line_debug_info): Use option variables.
4875         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
4876         (output_loc_list): Likewise.
4877         (add_high_low_attributes): Check option variables.
4878         Don't output entry view attribute in strict mode.
4879         (gen_inlined_subroutine_die): Check option variables.
4880         (dwarf2out_inline_entry): Likewise.
4881         (init_sections_and_labels): Likewise.
4882         (dwarf2out_early_finish): Likewise.
4883         (maybe_reset_location_view): New, from...
4884         (dwarf2out_var_location): ... here.  Call it.
4885         * debug.h (dwarf2out_default_as_loc_support): Declare.
4886         (dwarf2out_default_as_locview_support): Declare.
4887         * hooks.c (hook_int_rtx_insn_0): New.
4888         * hooks.h (hook_int_rtx_insn_0): Declare.
4889         * toplev.c (process_options): Take -gas-loc-support and
4890         -gas-locview-support from dwarf2out.  Enable
4891         -gvariable-location-views by default only with locview
4892         assembler support.  Enable -ginternal-reset-location-views by
4893         default only if the target defines the corresponding hook.
4894         Enable -ginline-points by default if location views are
4895         enabled; force it disabled if statement frontiers are
4896         disabled.
4897         * tree-inline.c (expand_call_inline): Check option variables.
4898         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
4899
4900 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
4901
4902         PR tree-optimization/84321
4903         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
4904         handling.  Also check whether the anti-range contains any values
4905         that satisfy the mask; switch to a VR_RANGE if not.
4906
4907 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
4908
4909         PR sanitizer/84340
4910         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
4911
4912 2018-02-13  Martin Jambor  <mjambor@suse.cz>
4913
4914         PR c++/83990
4915         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
4916         of call statements, also set location of a load to a temporary.
4917
4918 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
4919
4920         * config/rl78/rl78.c (add_vector_labels): New function.
4921         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
4922         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
4923         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
4924         which checks that no arguments are passed.
4925         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
4926         * doc/extend.texi: Documentation for the new attribute.
4927
4928 2018-02-13  Andreas Schwab  <schwab@suse.de>
4929
4930         * config/riscv/linux.h (CPP_SPEC): Define.
4931
4932 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4933
4934         PR target/84335
4935         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
4936         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
4937         OPTION_MASK_ISA_AES as first argument to def_builtin_const
4938         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
4939         instead of OPTION_MASK_ISA_PCLMUL as first argument to
4940         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
4941         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
4942         temporarily for AES and PCLMUL builtins.
4943
4944         PR tree-optimization/84339
4945         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
4946         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
4947         Formatting fixes.
4948
4949         PR middle-end/84309
4950         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
4951         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
4952         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
4953         inline function.
4954         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
4955         inline function.
4956         * omp-simd-clone.h: New file.
4957         * omp-simd-clone.c: Include omp-simd-clone.h.
4958         (expand_simd_clones): No longer static.
4959         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
4960         cgraph.h and omp-simd-clone.h.
4961         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
4962         (vect_recog_widen_shift_pattern): Formatting fix.
4963         (vect_pattern_recog_1): Don't check optab for calls.
4964
4965         PR target/84336
4966         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
4967         operands[2] into a REG before using gen_lowpart on it.
4968
4969 2018-02-12  Jeff Law  <law@redhat.com>
4970
4971         PR target/83760
4972         * config/sh/sh.c (find_barrier): Consider a sibling call
4973         a barrier as well.
4974
4975         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
4976         successfully back substituting a reg.
4977
4978 2018-02-12  Richard Biener  <rguenther@suse.de>
4979
4980         PR tree-optimization/84037
4981         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
4982         parameter, move visited init to caller.
4983         (vect_slp_analyze_operations): Separate cost from validity
4984         check, initialize visited once for all instances.
4985         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
4986         for all instances.
4987         * tree-vect-stmts.c (vect_model_simple_cost): Make early
4988         out an assert.
4989         (vect_model_promotion_demotion_cost): Likewise.
4990         (vectorizable_bswap): Guard cost modeling with !slp_node
4991         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
4992         SLP stmts.
4993         (vectorizable_call): Likewise.
4994         (vectorizable_conversion): Likewise.
4995         (vectorizable_assignment): Likewise.
4996         (vectorizable_shift): Likewise.
4997         (vectorizable_operation): Likewise.
4998         (vectorizable_store): Likewise.
4999         (vectorizable_load): Likewise.
5000         (vectorizable_condition): Likewise.
5001         (vectorizable_comparison): Likewise.
5002
5003 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
5004
5005         PR sanitizer/84307
5006         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
5007         (ASAN_MARK): Fix fnspec to account for return value, change pointer
5008         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
5009
5010 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
5011
5012         PR middle-end/83665
5013         * params.def (inline-min-speedup): Increase from 8 to 15.
5014         (max-inline-insns-auto): Decrease from 40 to 30.
5015         * ipa-split.c (consider_split): Add some buffer for function to
5016         be considered inlining candidate.
5017         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
5018         default values.
5019
5020 2018-02-12  Richard Biener  <rguenther@suse.de>
5021
5022         PR tree-optimization/84037
5023         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
5024         matched stmts if we cannot swap the non-matched ones.
5025
5026 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
5027
5028         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
5029         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
5030         _mm_maskz_scalef_round_ss): New intrinsics.
5031         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
5032         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
5033         __builtin_ia32_scalefss_round): Remove.
5034         (__builtin_ia32_scalefsd_mask_round,
5035         __builtin_ia32_scalefss_mask_round): New intrinsics.
5036         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
5037         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
5038         ((match_operand:VF_128 2 "<round_nimm_predicate>"
5039         "<round_constraint>")): Changed to ...
5040         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
5041         "<round_scalar_constraint>")): ... this.
5042         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
5043         %0, %1, %2<round_op3>}"): Changed to ...
5044         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
5045         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
5046         %2<round_scalar_mask_op3>}"): ... this.
5047         * config/i386/subst.md (round_scalar_nimm_predicate): New.
5048
5049 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
5050
5051         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
5052         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
5053         (_mm_maskz_sqrt_round_ss): New intrinsics.
5054         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
5055         (__builtin_ia32_sqrtsd_mask_round)
5056         (__builtin_ia32_sqrtss_mask_round): New builtins.
5057         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
5058         (__builtin_ia32_sqrtss_round): Remove.
5059         (__builtin_ia32_sqrtsd_mask_round)
5060         (__builtin_ia32_sqrtss_mask_round): New builtins.
5061         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
5062         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
5063         ((match_operand:VF_128 1 "vector_operand"
5064         "xBm,<round_constraint>")): Changed to ...
5065         ((match_operand:VF_128 1 "vector_operand"
5066         "xBm,<round_scalar_constraint>")): ... this.
5067         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
5068         %0, %2, %<iptr>1<round_op3>}): Changed to ...
5069         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
5070         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
5071         %<iptr>1<round_scalar_mask_op3>}): ... this.
5072         ((set_attr "prefix" "<round_prefix>")): Changed to ...
5073         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
5074
5075 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
5076
5077         PR target/84266
5078         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
5079         Cast vec_cmpeq result to correct type.
5080         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
5081         Cast vec_cmpgt result to correct type.
5082
5083 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
5084
5085         * final.c (final_scan_insn_1): Renamed from...
5086         (final_scan_insn): ... this.  New wrapper, to recover
5087         seen from the outermost call in recursive ones.
5088         * config/sparc/sparc.c (output_return): Drop seen from call.
5089         (output_sibcall): Likewise.
5090         * config/visium/visium.c (output_branch): Likewise.
5091
5092 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
5093
5094         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
5095         function label.
5096
5097 2018-02-10  Alan Modra  <amodra@gmail.com>
5098
5099         PR target/84300
5100         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
5101         Specify LR as an input.
5102
5103 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
5104
5105         PR sanitizer/83987
5106         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
5107         remove_member_access_dummy_vars): New functions.
5108         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
5109         lower_omp_1, execute_lower_omp): Use them.
5110
5111         PR rtl-optimization/84308
5112         * shrink-wrap.c (spread_components): Release todo vector.
5113
5114 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
5115
5116         PR rtl-optimization/57193
5117         * ira-color.c (struct allocno_color_data): Add member
5118         conflict_allocno_hard_prefs.
5119         (update_conflict_allocno_hard_prefs): New.
5120         (bucket_allocno_compare_func): Add a preference based on
5121         conflict_allocno_hard_prefs.
5122         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
5123         (color_allocnos): Remove a dead code.  Initiate
5124         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
5125
5126 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
5127
5128         PR target/84226
5129         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
5130         constraint from =wa to wa.  Avoid a subreg on the output operand,
5131         instead use a pseudo and subreg it in a move.
5132         (p9_xxbrd_<mode>): Changed to ...
5133         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
5134         (p9_xxbrd_v2df): New expander.
5135         (p9_xxbrw_<mode>): Changed to ...
5136         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
5137         (p9_xxbrw_v4sf): New expander.
5138
5139 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
5140
5141         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
5142
5143 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
5144
5145         PR target/83926
5146         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
5147         multiply in 32-bit mode.
5148         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
5149         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
5150         mode.
5151
5152 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
5153
5154         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
5155         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
5156         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
5157         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
5158
5159 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
5160
5161         PR lto/84213
5162         * dwarf2out.c (is_trivial_indirect_ref): New function.
5163         (dwarf2out_late_global_decl): Do not generate a location
5164         attribute for variables that have a non-trivial DECL_VALUE_EXPR
5165         and that are not defined in the current unit.
5166
5167 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
5168
5169         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
5170         instead of a libcall for UNORDERED.
5171
5172 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
5173
5174         PR target/82641
5175         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
5176         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
5177
5178 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5179
5180         PR target/PR84295
5181         * config/s390/s390.c (s390_set_current_function): Invoke
5182         s390_indirect_branch_settings also if fndecl didn't change.
5183
5184 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
5185
5186         * config/rs6000/rs6000.md (blockage): Set length to zero.
5187
5188 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
5189
5190         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
5191
5192 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
5193
5194         PR sanitizer/84285
5195         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
5196         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
5197         -static-lib*san.
5198
5199         PR debug/84252
5200         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
5201         PARALLEL incoming that failed vt_get_decl_and_offset check.
5202
5203         PR middle-end/84237
5204         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
5205         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
5206         TREE_READONLY bit.
5207         (get_variable_section): For decls in named .bss* sections pass true as
5208         second argument to bss_initializer_p.
5209
5210 2018-02-09  Marek Polacek  <polacek@redhat.com>
5211             Jakub Jelinek  <jakub@redhat.com>
5212
5213         PR c++/83659
5214         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
5215         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
5216         Sync some changes from cxx_fold_indirect_ref.
5217
5218 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
5219
5220         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
5221         markers.
5222         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
5223         (BLOCK_INLINE_ENTRY_LABEL): New.
5224         (dwarf2out_var_location): Disregard inline entry markers.
5225         (inline_entry_data): New struct.
5226         (inline_entry_data_hasher): New hashtable type.
5227         (inline_entry_data_hasher::hash): New.
5228         (inline_entry_data_hasher::equal): New.
5229         (inline_entry_data_table): New variable.
5230         (add_high_low_attributes): Add DW_AT_entry_pc and
5231         DW_AT_GNU_entry_view attributes if a pending entry is found
5232         in inline_entry_data_table.  Add old entry_pc attribute only
5233         if debug nonbinding markers are disabled.
5234         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
5235         markers are enabled.
5236         (block_within_block_p, dwarf2out_inline_entry): New.
5237         (dwarf2out_finish): Check that no entries remained in
5238         inline_entry_data_table.
5239         * final.c (reemit_insn_block_notes): Handle inline entry notes.
5240         (final_scan_insn, notice_source_line): Likewise.
5241         (rest_of_clean_state): Skip inline entry markers.
5242         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
5243         markers.
5244         * gimple.c (gimple_build_debug_inline_entry): New.
5245         * gimple.h (enum gimple_debug_subcode): Add
5246         GIMPLE_DEBUG_INLINE_ENTRY.
5247         (gimple_build_debug_inline_entry): Declare.
5248         (gimple_debug_inline_entry_p): New.
5249         (gimple_debug_nonbind_marker_p): Adjust.
5250         * insn-notes.def (INLINE_ENTRY): New.
5251         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
5252         inline entry marker notes.
5253         (print_insn): Likewise.
5254         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
5255         (INSN_DEBUG_MARKER_KIND): Likewise.
5256         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
5257         * tree-inline.c (expand_call_inline): Build and insert
5258         debug_inline_entry stmt.
5259         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
5260         inline entry blocks early, if nonbind markers are enabled.
5261         (dump_scope_block): Dump fragment info.
5262         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
5263         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
5264         (gimple_build_debug_inline_entry): New.
5265         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
5266         Enable/disable inline entry points too.
5267         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
5268         (DEBUG_INSN): Describe inline entry markers.
5269
5270         * common.opt (gvariable-location-views): New.
5271         (gvariable-location-views=incompat5): New.
5272         * config.in: Rebuilt.
5273         * configure: Rebuilt.
5274         * configure.ac: Test assembler for view support.
5275         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
5276         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
5277         * dwarf2out.c (var_loc_view): New typedef.
5278         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
5279         (dwarf2out_locviews_in_attribute): New.
5280         (dwarf2out_locviews_in_loclist): New.
5281         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
5282         (enum dw_line_info_opcode): Add LI_adv_address.
5283         (struct dw_line_info_table): Add view.
5284         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
5285         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
5286         (zero_view_p): New variable.
5287         (ZERO_VIEW_P): New macro.
5288         (output_asm_line_debug_info): New.
5289         (struct var_loc_node): Add view.
5290         (add_AT_view_list, AT_loc_list): New.
5291         (add_var_loc_to_decl): Add view param.  Test it against last.
5292         (new_loc_list): Add view params.  Record them.
5293         (AT_loc_list_ptr): Handle loc and view lists.
5294         (view_list_to_loc_list_val_node): New.
5295         (print_dw_val): Handle dw_val_class_view_list.
5296         (size_of_die): Likewise.
5297         (value_format): Likewise.
5298         (loc_list_has_views): New.
5299         (gen_llsym): Set vl_symbol too.
5300         (maybe_gen_llsym, skip_loc_list_entry): New.
5301         (dwarf2out_maybe_output_loclist_view_pair): New.
5302         (output_loc_list): Output view list or entries too.
5303         (output_view_list_offset): New.
5304         (output_die): Handle dw_val_class_view_list.
5305         (output_dwarf_version): New.
5306         (output_compilation_unit_header): Use it.
5307         (output_skeleton_debug_sections): Likewise.
5308         (output_rnglists, output_line_info): Likewise.
5309         (output_pubnames, output_aranges): Update version comments.
5310         (output_one_line_info_table): Output view numbers in asm comments.
5311         (dw_loc_list): Determine current endview, pass it to new_loc_list.
5312         Call maybe_gen_llsym.
5313         (loc_list_from_tree_1): Adjust.
5314         (add_AT_location_description): Create view list attribute if
5315         needed, check it's absent otherwise.
5316         (convert_cfa_to_fb_loc_list): Adjust.
5317         (maybe_emit_file): Call output_asm_line_debug_info for test.
5318         (dwarf2out_var_location): Reset views as needed.  Precompute
5319         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
5320         attribute.  Set view.
5321         (new_line_info_table): Reset next view.
5322         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
5323         (dwarf2out_source_line): Likewise.  Output view resets and labels to
5324         the assembler, or select appropriate line info opcodes.
5325         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
5326         (optimize_string_length): Catch it.  Adjust.
5327         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
5328         dw_val_class_view_list, and remove it if no longer needed.
5329         (hash_loc_list): Hash view numbers.
5330         (loc_list_hasher::equal): Compare them.
5331         (optimize_location_lists): Check whether a view list symbol is
5332         needed, and whether the locview attribute is present, and
5333         whether they match.  Remove the locview attribute if no longer
5334         needed.
5335         (index_location_lists): Call skip_loc_list_entry for test.
5336         (dwarf2out_finish): Call output_asm_line_debug_info for test.
5337         Use output_dwarf_version.
5338         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
5339         (struct dw_val_node): Add val_view_list.
5340         * final.c (SEEN_NEXT_VIEW): New.
5341         (set_next_view_needed): New.
5342         (clear_next_view_needed): New.
5343         (maybe_output_next_view): New.
5344         (final_start_function): Rename to...
5345         (final_start_function_1): ... this.  Take pointer to FIRST,
5346         add SEEN parameter.  Emit param bindings in the initial view.
5347         (final_start_function): Reintroduce SEEN-less interface.
5348         (final): Rename to...
5349         (final_1): ... this.  Take SEEN parameter.  Output final pending
5350         next view at the end.
5351         (final): Reintroduce seen-less interface.
5352         (final_scan_insn): Output pending next view before switching
5353         sections or ending a block.  Mark the next view as needed when
5354         outputting variable locations.  Notify debug backend of section
5355         changes, and of location view changes.
5356         (rest_of_handle_final): Adjust.
5357         * toplev.c (process_options): Autodetect value for debug variable
5358         location views option.  Warn on incompat5 without -gdwarf-5.
5359         * doc/invoke.texi (gvariable-location-views): New.
5360         (gvariable-location-views=incompat5): New.
5361         (gno-variable-location-views): New.
5362
5363 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
5364
5365         PR tree-optimization/84136
5366         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
5367         that the result of find_edge is non-NULL.
5368
5369 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
5370
5371         PR target/83008
5372         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
5373         storing integer register in SImode.  Fix cost of 256 and 512
5374         byte aligned SSE register store.
5375
5376 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
5377
5378         * config/i386/i386.c (ix86_multiplication_cost): Fix
5379         multiplication cost for TARGET_AVX512DQ.
5380
5381 2018-02-08  Marek Polacek  <polacek@redhat.com>
5382
5383         PR tree-optimization/84238
5384         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
5385         get_range_strlen.
5386
5387 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5388
5389         PR tree-optimization/84265
5390         * tree-vect-stmts.c (vectorizable_store): Don't treat
5391         VMAT_CONTIGUOUS accesses as grouped.
5392         (vectorizable_load): Likewise.
5393
5394 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5395
5396         PR tree-optimization/81635
5397         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
5398         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
5399         (test_round_for_mask): New functions.
5400         (wide_int_cc_tests): Call test_round_for_mask.
5401         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
5402         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
5403         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
5404         range returned by get_range_info.
5405
5406 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
5407
5408         PR ipa/81360
5409         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
5410         * symtab.c: Include builtins.h
5411         (symtab_node::output_to_lto_symbol_table_p): Move here
5412         from lto-streamer-out.c:output_symbol_p.
5413         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
5414         (output_symbol_p): Move all logic to symtab.c
5415         (produce_symtab): Update.
5416
5417 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5418
5419         * config/s390/s390-opts.h (enum indirect_branch): Define.
5420         * config/s390/s390-protos.h (s390_return_addr_from_memory)
5421         (s390_indirect_branch_via_thunk)
5422         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
5423         (enum s390_indirect_branch_type): Define.
5424         * config/s390/s390.c (struct s390_frame_layout, struct
5425         machine_function): Remove.
5426         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
5427         (indirect_branch_table_label_no, indirect_branch_table_name):
5428         Define variables.
5429         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
5430         (enum s390_indirect_branch_option): Define.
5431         (s390_return_addr_from_memory): New function.
5432         (s390_handle_string_attribute): New function.
5433         (s390_attribute_table): Add new attribute handler.
5434         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
5435         (s390_indirect_branch_via_thunk): New function.
5436         (s390_indirect_branch_via_inline_thunk): New function.
5437         (s390_function_ok_for_sibcall): When jumping via thunk disallow
5438         sibling call optimization for non z10 compiles.
5439         (s390_emit_call): Force indirect branch target to be a single
5440         register.  Add r1 clobber for non-z10 compiles.
5441         (s390_emit_epilogue): Emit return jump via return_use expander.
5442         (s390_reorg): Handle JUMP_INSNs as execute targets.
5443         (s390_option_override_internal): Perform validity checks for the
5444         new command line options.
5445         (s390_indirect_branch_attrvalue): New function.
5446         (s390_indirect_branch_settings): New function.
5447         (s390_set_current_function): Invoke s390_indirect_branch_settings.
5448         (s390_output_indirect_thunk_function):  New function.
5449         (s390_code_end): Implement target hook.
5450         (s390_case_values_threshold): Implement target hook.
5451         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
5452         macros.
5453         * config/s390/s390.h (struct s390_frame_layout)
5454         (struct machine_function): Move here from s390.c.
5455         (TARGET_INDIRECT_BRANCH_NOBP_RET)
5456         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
5457         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
5458         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
5459         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
5460         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
5461         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
5462         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
5463         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
5464         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
5465         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
5466         (mnemonic attribute): Add values which aren't recognized
5467         automatically.
5468         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
5469         pattern for branch conversion.  Fix mnemonic attribute.
5470         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
5471         indirect branch via thunk if requested.
5472         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
5473         ("*indirect_jump"): Disable for branch conversion using out of
5474         line thunks.
5475         ("indirect_jump_via_thunk<mode>_z10")
5476         ("indirect_jump_via_thunk<mode>")
5477         ("indirect_jump_via_inlinethunk<mode>_z10")
5478         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
5479         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
5480         ("casesi_jump_via_inlinethunk<mode>_z10")
5481         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
5482         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
5483         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
5484         ("*indirect2_jump"): Disable for branch conversion.
5485         ("casesi_jump"): Turn into expander and expand patterns for branch
5486         conversion.
5487         ("return_use"): New expander.
5488         ("*return"): Emit return via thunk and rename it to ...
5489         ("*return<mode>"): ... this one.
5490         * config/s390/s390.opt: Add new options and and enum for the
5491         option values.
5492
5493 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5494
5495         * lra-constraints.c (match_reload): Unconditionally use
5496         gen_lowpart_SUBREG, rather than selecting between that
5497         and equivalent gen_rtx_SUBREG code.
5498
5499 2018-02-08  Richard Biener  <rguenther@suse.de>
5500
5501         PR tree-optimization/84233
5502         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
5503         changed flag instead of boguously re-using phi_inserted.
5504
5505 2018-02-08  Martin Jambor  <mjambor@suse.cz>
5506
5507         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
5508         static local variables.
5509
5510 2018-02-08  Richard Biener  <rguenther@suse.de>
5511
5512         PR tree-optimization/84278
5513         * tree-vect-stmts.c (vectorizable_store): When looking for
5514         smaller vector types to perform grouped strided loads/stores
5515         make sure the mode is supported by the target.
5516         (vectorizable_load): Likewise.
5517
5518 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5519
5520         * config/aarch64/aarch64.c (aarch64_components_for_bb):
5521         Increase LDP/STP opportunities by adding adjacent callee-saves.
5522
5523 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5524
5525         PR rtl-optimization/84068
5526         PR rtl-optimization/83459
5527         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
5528
5529 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
5530
5531         PR tree-optimization/84224
5532         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
5533         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
5534         non-zero arguments.
5535
5536 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
5537
5538         PR target/84113
5539         * config/rs6000/altivec.md (*restore_world): Remove LR use.
5540         * config/rs6000/predicates.md (restore_world_operation): Adjust op
5541         count, remove one USE.
5542
5543 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
5544
5545         * doc/install.texi (Configuration): Document the
5546         --with-long-double-format={ibm,ieee} PowerPC configuration
5547         options.
5548
5549         PR target/84154
5550         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
5551         Convert from define_expand to be define_insn_and_split.  Rework
5552         float/double/_Float128 conversions to QI/HI/SImode to work with
5553         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
5554         conversions to QI/HImode types did a store and then a load to
5555         truncate the value.  For conversions to VSX registers, don't split
5556         the insn, instead emit the code directly.  Use the code iterator
5557         any_fix to combine signed and unsigned conversions.
5558         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
5559         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
5560         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5561         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
5562         (fix_<mode>di2_hw): Likewise.
5563         (fixuns_<mode>di2_hw): Likewise.
5564         (fix_<mode>si2_hw): Likewise.
5565         (fixuns_<mode>si2_hw): Likewise.
5566         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
5567         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5568         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
5569         fix<uns>_trunc<SFDF:mode>si2_p8.
5570         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
5571         used.
5572         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
5573         (fix<uns>_<mode>_mem): Likewise.
5574         (fctiw<u>z_<mode>_mem): Likewise.
5575         (fix<uns>_<mode>_mem): Likewise.
5576         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
5577         the register allocator from doing a direct move to the GPRs to do
5578         a store, and instead use the ISA 3.0 store byte/half-word from
5579         vector register instruction.  For IEEE 128-bit floating point,
5580         also optimize stores of 32-bit ints.
5581         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
5582
5583 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
5584
5585         * genextract.c (push_pathstr_operand): New function to support
5586         [a-zA-Z].
5587         (walk_rtx): Call push_pathstr_operand.
5588         (print_path): Support [a-zA-Z].
5589
5590 2018-02-07  Richard Biener  <rguenther@suse.de>
5591
5592         PR tree-optimization/84037
5593         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
5594         (cse_and_gimplify_to_preheader): Declare.
5595         (vect_get_place_in_interleaving_chain): Likewise.
5596         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
5597         ivexpr_map.
5598         (_loop_vec_info::~_loop_vec_info): Delete it.
5599         (cse_and_gimplify_to_preheader): New function.
5600         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
5601         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
5602         (vectorizable_load): Likewise.  For grouped stores always base
5603         the IV on the first element.
5604         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
5605         condition before gimplifying.
5606
5607 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5608
5609         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
5610         *DIV_EXPR and *MOD_EXPR.
5611
5612 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
5613
5614         PR target/84248
5615         * config/i386/i386.c (ix86_option_override_internal): Mask out
5616         the CF_SET bit when checking -fcf-protection.
5617
5618 2018-02-07  Tom de Vries  <tom@codesourcery.com>
5619
5620         PR libgomp/84217
5621         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
5622         enough.
5623
5624 2018-02-07  Richard Biener  <rguenther@suse.de>
5625
5626         PR tree-optimization/84204
5627         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
5628         this place.
5629
5630         PR tree-optimization/84205
5631         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
5632         special-case isl_ast_op_zdiv_r.
5633
5634         PR tree-optimization/84223
5635         * graphite-scop-detection.c (gather_bbs::before_dom_children):
5636         Only add conditions from within the region.
5637         (gather_bbs::after_dom_children): Adjust.
5638
5639 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
5640
5641         PR target/84209
5642         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
5643         * config/avr/avr.md: Only post-reload split REG-REG moves if
5644         either register is GENERAL_REG_P.
5645
5646 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5647
5648         PR tree-optimization/84235
5649         * tree-ssa-scopedtables.c
5650         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
5651         if the subtraction is performed in floating point type where NaNs are
5652         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
5653         build 1.  Formatting fix.
5654
5655 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
5656
5657         PR target/84146
5658         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
5659         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
5660         and skip it regardless of bb boundaries.  Use CALL_P macro,
5661         don't test INSN_P (insn) together with CALL_P or JUMP_P check
5662         unnecessarily, formatting fix.
5663
5664 2018-02-06  Michael Collison  <michael.collison@arm.com>
5665
5666         * config/arm/thumb2.md:
5667         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
5668         (*thumb_mov_notscc): Ditto.
5669
5670 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
5671
5672         PR target/84154
5673         * config/rs6000/rs6000.md (su code attribute): Use "u" for
5674         unsigned_fix, not "s".
5675
5676 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5677
5678         * configure.ac (gcc_fn_eh_frame_ro): New function.
5679         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
5680         correct .eh_frame permissions.
5681         * configure: Regenerate.
5682
5683 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
5684
5685         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
5686         irrelevant options.
5687
5688 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5689
5690         * config/rs6000/rs6000.c (rs6000_option_override_internal):
5691         Display warning message for -mno-speculate-indirect-jumps.
5692
5693 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
5694
5695         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
5696         Undocumented.
5697         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
5698
5699 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
5700
5701         PR tree-optimization/84225
5702         * tree-eh.c (find_trapping_overflow): Only call
5703         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
5704
5705 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5706
5707         PR target/84145
5708         * config/i386/i386.c: Reimplement the check of possible options
5709         -mibt/-mshstk conbination. Change error messages.
5710         * doc/invoke.texi: Fix a typo: remove extra '='.
5711
5712 2018-02-06  Marek Polacek  <polacek@redhat.com>
5713
5714         PR tree-optimization/84228
5715         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
5716
5717 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
5718
5719         PR target/82641
5720         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
5721         emitted arch directives.
5722         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
5723         __ARM_FEATURE_COPROC before changing architectures.
5724
5725 2018-02-06  Richard Biener  <rguenther@suse.de>
5726
5727         * config/i386/i386.c (print_reg): Fix typo.
5728         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
5729
5730 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
5731
5732         * configure: Regenerate.
5733
5734 2018-02-05  Martin Sebor  <msebor@redhat.com>
5735
5736         PR tree-optimization/83369
5737         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
5738         inlining context.
5739
5740 2018-02-05  Martin Liska  <mliska@suse.cz>
5741
5742         * doc/invoke.texi: Cherry-pick upstream r323995.
5743
5744 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
5745
5746         * ira.c (ira_init_register_move_cost): Adjust comment.
5747
5748 2018-02-05  Martin Liska  <mliska@suse.cz>
5749
5750         PR gcov-profile/84137
5751         * doc/gcov.texi: Fix typo in documentation.
5752
5753 2018-02-05  Martin Liska  <mliska@suse.cz>
5754
5755         PR gcov-profile/83879
5756         * doc/gcov.texi: Document necessity of --dynamic-list-data when
5757         using dlopen functionality.
5758
5759 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
5760
5761         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
5762         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
5763         _mm_maskz_range_ss, _mm_mask_range_round_ss,
5764         _mm_maskz_range_round_ss): New intrinsics.
5765         (__builtin_ia32_rangesd128_round)
5766         (__builtin_ia32_rangess128_round): Remove.
5767         (__builtin_ia32_rangesd128_mask_round,
5768         __builtin_ia32_rangess128_mask_round): New builtins.
5769         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
5770         __builtin_ia32_rangess128_round): Remove.
5771         (__builtin_ia32_rangesd128_mask_round,
5772         __builtin_ia32_rangess128_mask_round): New builtins.
5773         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
5774         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
5775         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
5776         "<round_saeonly_constraint>")): Changed to ...
5777         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
5778         "<round_saeonly_scalar_constraint>")): ... this.
5779         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
5780         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
5781         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
5782         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
5783         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
5784
5785 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
5786
5787         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
5788         options.
5789         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
5790         Remove all values except native, 8540 and 8548.
5791
5792 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
5793
5794         * config/i386/i386.c (ix86_output_function_return): Pass
5795         INVALID_REGNUM, instead of -1, as invalid register number to
5796         indirect_thunk_name and output_indirect_thunk.
5797
5798 2018-02-02  Julia Koval  <julia.koval@intel.com>
5799
5800         * config.gcc: Add -march=icelake.
5801         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
5802         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
5803         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
5804         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
5805         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
5806         (processor_target_table): Add icelake.
5807         (ix86_option_override_internal): Handle new PTAs.
5808         (get_builtin_code_for_version): Handle icelake.
5809         (M_INTEL_COREI7_ICELAKE): New.
5810         (fold_builtin_cpu): Handle icelake.
5811         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
5812         * doc/invoke.texi: Add -march=icelake.
5813
5814 2018-02-02  Julia Koval  <julia.koval@intel.com>
5815
5816         * config/i386/i386.c (ix86_option_override_internal): Change flags type
5817         to wide_int_bitmask.
5818         * wide-int-bitmask.h: New.
5819
5820 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5821
5822         PR target/84066
5823         * config/i386/i386.md: Replace Pmode with word_mode in
5824         builtin_setjmp_setup and builtin_longjmp to support x32.
5825
5826 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
5827
5828         PR target/56010
5829         PR target/83743
5830         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
5831         #include "opts.h".
5832         (rs6000_supported_cpu_names): New static variable.
5833         (linux_cpu_translation_table): Likewise.
5834         (elf_platform) <cpu>: Define new static variable and use it.
5835         Translate kernel AT_PLATFORM name to canonical name if needed.
5836         Error if platform name is unknown.
5837
5838 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
5839
5840         PR target/84089
5841         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
5842
5843 2018-02-01  Jeff Law  <law@redhat.com>
5844
5845         PR target/84128
5846         * config/i386/i386.c (release_scratch_register_on_entry): Add new
5847         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
5848         the scratch if RELEASE_VIA_POP is false.
5849         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
5850         If we have to save a temporary register, decrement SIZE appropriately.
5851         Pass new arguments to release_scratch_register_on_entry.
5852         (ix86_adjust_stack_and_probe): Likewise.
5853         (ix86_emit_probe_stack_range): Pass new arguments to
5854         release_scratch_register_on_entry.
5855
5856 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
5857
5858         PR rtl-optimization/84157
5859         * combine.c (change_zero_ext): Use REG_P predicate in
5860         front of HARD_REGISTER_P predicate.
5861
5862 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
5863
5864         * config/avr/avr.c (avr_option_override): Move disabling of
5865         -fdelete-null-pointer-checks to...
5866         * common/config/avr/avr-common.c (avr_option_optimization_table):
5867         ...here.
5868
5869 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5870
5871         PR tree-optimization/81635
5872         * tree-data-ref.c (split_constant_offset_1): For types that
5873         wrap on overflow, try to use range info to prove that wrapping
5874         cannot occur.
5875
5876 2018-02-01  Renlin Li  <renlin.li@arm.com>
5877
5878         PR target/83370
5879         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
5880         TAILCALL_ADDR_REGS.
5881         (aarch64_register_move_cost): Likewise.
5882         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
5883         TAILCALL_ADDR_REGS.
5884         (REG_CLASS_NAMES): Likewise.
5885         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
5886         TAILCALL_ADDR_REGS. Remove IP registers.
5887         * config/aarch64/aarch64.md (Ucs): Update register constraint.
5888
5889 2018-02-01  Richard Biener  <rguenther@suse.de>
5890
5891         * domwalk.h (dom_walker::dom_walker): Add additional constructor
5892         for specifying RPO order and allow NULL for that.
5893         * domwalk.c (dom_walker::dom_walker): Likewise.
5894         (dom_walker::walk): Handle NULL RPO order.
5895         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
5896         in RPO order.
5897         (rewrite_update_dom_walker): Likewise.
5898         (mark_def_dom_walker): Likewise.
5899
5900 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5901
5902         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
5903         (aarch64_maybe_expand_sve_subreg_move): Declare.
5904         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
5905         * config/aarch64/predicates.md (aarch64_any_register_operand): New
5906         predicate.
5907         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
5908         that are semantically a reverse operation.
5909         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
5910         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
5911         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
5912         functions.
5913         (aarch64_can_change_mode_class): For big-endian, forbid changes
5914         between two SVE modes if they have different element sizes.
5915
5916 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5917
5918         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
5919         the TImode handling for big-endian targets.
5920
5921 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5922
5923         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
5924         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
5925         not just bytes.
5926         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
5927         Remove BSWAP handing for big-endian targets and use the form of
5928         LD1RQ appropariate for the mode.
5929
5930 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5931
5932         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
5933         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
5934         duplicated element.
5935
5936 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5937
5938         PR tearget/83845
5939         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
5940         check for operands that need to go through aarch64_sve_reload_be.
5941
5942 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
5943
5944         PR tree-optimization/81661
5945         PR tree-optimization/84117
5946         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
5947         * tree-eh.c: Include gimplify.h.
5948         (find_trapping_overflow, replace_trapping_overflow,
5949         rewrite_to_non_trapping_overflow): New functions.
5950         * tree-vect-loop.c: Include tree-eh.h.
5951         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
5952         * tree-data-ref.c: Include tree-eh.h.
5953         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
5954
5955 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
5956
5957         PR rtl-optimization/84123
5958         * combine.c (change_zero_ext): Check if hard register satisfies
5959         can_change_dest_mode before calling gen_lowpart_SUBREG.
5960
5961 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
5962
5963         PR target/82444
5964         * ira.c (ira_init_register_move_cost): Remove assert.
5965
5966 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5967
5968         PR rtl-optimization/84071
5969         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
5970         * doc/tm.texi: Regenerate.
5971
5972 2018-01-31  Richard Biener  <rguenther@suse.de>
5973
5974         PR tree-optimization/84132
5975         * tree-data-ref.c (analyze_miv_subscript): Properly
5976         check whether evolution_function_is_affine_multivariate_p
5977         before calling gcd_of_steps_may_divide_p.
5978
5979 2018-01-31  Julia Koval  <julia.koval@intel.com>
5980
5981         PR target/83618
5982         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
5983         * config/i386/i386.md (rdpid_rex64) New.
5984         (rdpid): Make 32bit only.
5985
5986 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
5987
5988         PR lto/84105
5989         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
5990         an IDENTIFIER_NODE for FUNCTION_TYPE's.
5991
5992 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5993
5994         Revert
5995         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
5996
5997         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5998
5999 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
6000
6001         PR rtl-optimization/84071
6002         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
6003         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
6004
6005 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
6006
6007         * config/arc/arc.c (arc_handle_aux_attribute): New function.
6008         (arc_attribute_table): Add 'aux' attribute.
6009         (arc_in_small_data_p): Consider aux like variables.
6010         (arc_is_aux_reg_p): New function.
6011         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
6012         (arc_get_aux_arg): New function.
6013         (prepare_move_operands): Handle aux-register access.
6014         (arc_handle_aux_attribute): New function.
6015         * doc/extend.texi (ARC Variable attributes): Add subsection.
6016
6017 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
6018
6019         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
6020         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
6021         (arc_attribute_table): Add 'uncached' attribute.
6022         (arc_print_operand): Print '.di' flag for uncached memory
6023         accesses.
6024         (arc_in_small_data_p): Do not consider for small data the uncached
6025         types.
6026         (arc_is_uncached_mem_p): New function.
6027         * config/arc/predicates.md (compact_store_memory_operand): Check
6028         for uncached memory accesses.
6029         (nonvol_nonimm_operand): Likewise.
6030         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
6031
6032 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
6033
6034         PR c/84100
6035         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
6036         falign-loops=): Add Optimization flag.
6037
6038 2018-01-30  Jeff Law  <law@redhat.com>
6039
6040         PR target/84064
6041         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
6042         INT_REGISTERS_SAVED.  Check it prior to calling
6043         get_scratch_register_on_entry.
6044         (ix86_adjust_stack_and_probe): Similarly.
6045         (ix86_emit_probe_stack_range): Similarly.
6046         (ix86_expand_prologue): Corresponding changes.
6047
6048 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6049
6050         PR target/40411
6051         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
6052         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
6053
6054 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
6055
6056         PR target/84112
6057         * lra-constraints.c (curr_insn_transform): Process AND in the
6058         address.
6059
6060 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
6061
6062         PR rtl-optimization/83986
6063         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
6064         dependence against last_pending_memory_flush in addition to
6065         pending_jump_insns.
6066
6067 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
6068
6069         PR tree-optimization/81611
6070         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
6071         copies.
6072
6073 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
6074
6075         PR target/83758
6076         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
6077         a reg rtx.
6078
6079 2018-01-30  Richard Biener  <rguenther@suse.de>
6080             Jakub Jelinek  <jakub@redhat.com>
6081
6082         PR tree-optimization/84111
6083         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
6084         inner loops added during recursion, as they don't have up-to-date
6085         SSA form.
6086
6087 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
6088
6089         PR ipa/81360
6090         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
6091         (can_inline_edge_by_limits_p): ... here.
6092         (can_early_inline_edge_p, check_callers,
6093         update_caller_keys, update_callee_keys, recursive_inlining,
6094         add_new_edges_to_heap, speculation_useful_p,
6095         inline_small_functions,
6096         inline_small_functions, flatten_function,
6097         inline_to_all_callers_1): Update.
6098
6099 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
6100
6101         * profile-count.c (profile_count::combine_with_ipa_count): Handle
6102         zeros correctly.
6103
6104 2018-01-30  Richard Biener  <rguenther@suse.de>
6105
6106         PR tree-optimization/83008
6107         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
6108         invariant and constant vector uses in stmts when they need
6109         more than one stmt.
6110
6111 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6112
6113         PR bootstrap/84017
6114         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
6115         * configure: Regenerate.
6116
6117 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6118
6119         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
6120         pattern.
6121         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
6122         Use gen_rtx_REG rather than gen_lowpart.
6123
6124 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6125
6126         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
6127         rather than 0 when creating partial subregs.
6128
6129 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6130
6131         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
6132         of usage.
6133
6134 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
6135
6136         PR target/81550
6137         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
6138         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
6139         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
6140         flags.  This restores the settings used before the 2017-07-24.
6141         Turning off pre increment/decrement/modify allows IVOPTS to
6142         optimize DF/SF loops where the index is an int.
6143
6144 2018-01-29  Richard Biener  <rguenther@suse.de>
6145             Kelvin Nilsen  <kelvin@gcc.gnu.org>
6146
6147         PR bootstrap/80867
6148         * tree-vect-stmts.c (vectorizable_call): Don't call
6149         targetm.vectorize_builtin_md_vectorized_function if callee is
6150         NULL.
6151
6152 2018-01-22  Carl Love  <cel@us.ibm.com>
6153
6154         * doc/extend.tex: Fix typo in second arg in
6155         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
6156
6157 2018-01-29  Richard Biener  <rguenther@suse.de>
6158
6159         PR tree-optimization/84086
6160         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
6161         (flush_ssaname_freelist): When SSA names were released reset
6162         the SCEV hash table.
6163
6164 2018-01-29  Richard Biener  <rguenther@suse.de>
6165
6166         PR tree-optimization/84057
6167         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
6168         removed paths when removing edges.
6169
6170 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
6171
6172         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
6173         -mfunction-return=@var{choice}.
6174
6175 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
6176
6177         PR diagnostic/84034
6178         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
6179         Handle CR like TAB.
6180         (layout::print_source_line): Likewise.
6181         (test_get_line_width_without_trailing_whitespace): Add test cases.
6182
6183 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
6184
6185         PR middle-end/84040
6186         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
6187         debug insns.
6188
6189 2018-01-26  Jim Wilson  <jimw@sifive.com>
6190
6191         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
6192
6193         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
6194         specified.
6195
6196 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6197
6198         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
6199         and CMP + SUB-immediate -> SUBS.
6200
6201 2018-01-26  Martin Sebor  <msebor@redhat.com>
6202
6203         PR tree-optimization/83896
6204         * tree-ssa-strlen.c (get_string_len): Rename...
6205         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
6206         Avoid assuming length is constant.
6207         (handle_char_store): Use HOST_WIDE_INT for string length.
6208
6209 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
6210
6211         PR target/81763
6212         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
6213         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
6214
6215 2018-01-26  Richard Biener  <rguenther@suse.de>
6216
6217         PR rtl-optimization/84003
6218         * dse.c (record_store): Only record redundant stores when
6219         the earlier store aliases at least all accesses the later one does.
6220
6221 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
6222
6223         PR rtl-optimization/83985
6224         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
6225         REG_CFA_RESTORE insns.
6226         (delete_unmarked_insns): Don't ignore separate shrink wrapping
6227         REG_CFA_RESTORE insns here.
6228
6229         PR c/83989
6230         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
6231         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
6232
6233 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6234
6235         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
6236         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
6237         (arc_init): Likewise.
6238         (arc_override_options): Likewise.
6239         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
6240         value.
6241         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
6242         support.
6243         * config/arc/arc.h (TARGET_DBNZ): Define.
6244         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
6245         properly set the tune attribute.
6246         (dbnz): Use TARGET_DBNZ guard.
6247         * config/arc/arc.opt (mtune): Add core3 option.
6248
6249 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6250
6251         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
6252         recognize new pic like addresses.
6253         (arc_delegitimize_address): Clean up.
6254
6255 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6256
6257         * config/arc/arc-arches.def: Option mrf16 valid for all
6258         architectures.
6259         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
6260         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
6261         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
6262         * config/arc/arc-tables.opt: Regenerate.
6263         * config/arc/arc.c (arc_conditional_register_usage): Handle
6264         reduced register file case.
6265         (arc_file_start): Set must have build attributes.
6266         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
6267         mrf16 option value.
6268         * config/arc/arc.opt (mrf16): Add new option.
6269         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
6270         * config/arc/genmultilib.awk: Handle new mrf16 option.
6271         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
6272         * config/arc/t-multilib: Regenerate.
6273         * doc/invoke.texi (ARC Options): Document mrf16 option.
6274
6275 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6276
6277         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
6278         * config/arc/arc.c (arc_handle_secure_attribute): New function.
6279         (arc_attribute_table): Add 'secure_call' attribute.
6280         (arc_print_operand): Print secure call operand.
6281         (arc_function_ok_for_sibcall): Don't optimize tail calls when
6282         secure.
6283         (arc_is_secure_call_p): New function.  * config/arc/arc.md
6284         (call_i): Add support for sjli instruction.
6285         (call_value_i): Likewise.
6286         * config/arc/constraints.md (Csc): New constraint.
6287
6288 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6289             John Eric Martin  <John.Martin@emmicro-us.com>
6290
6291         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
6292         * config/arc/arc.c (_arc_jli_section): New struct.
6293         (arc_jli_section): New type.
6294         (rc_jli_sections): New static variable.
6295         (arc_handle_jli_attribute): New function.
6296         (arc_attribute_table): Add jli_always and jli_fixed attribute.
6297         (arc_file_end): New function.
6298         (TARGET_ASM_FILE_END): Define.
6299         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
6300         (arc_add_jli_section): New function.
6301         (jli_call_scan): Likewise.
6302         (arc_reorg): Call jli_call_scan.
6303         (arc_output_addsi): Remove 'S' from printing asm operand.
6304         (arc_is_jli_call_p): New function.
6305         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
6306         operand.
6307         (movhi_insn): Likewise.
6308         (movsi_insn): Likewise.
6309         (movsi_set_cc_insn): Likewise.
6310         (loadqi_update): Likewise.
6311         (load_zeroextendqisi_update): Likewise.
6312         (load_signextendqisi_update): Likewise.
6313         (loadhi_update): Likewise.
6314         (load_zeroextendhisi_update): Likewise.
6315         (load_signextendhisi_update): Likewise.
6316         (loadsi_update): Likewise.
6317         (loadsf_update): Likewise.
6318         (movsicc_insn): Likewise.
6319         (bset_insn): Likewise.
6320         (bxor_insn): Likewise.
6321         (bclr_insn): Likewise.
6322         (bmsk_insn): Likewise.
6323         (bicsi3_insn): Likewise.
6324         (cmpsi_cc_c_insn): Likewise.
6325         (movsi_ne): Likewise.
6326         (movsi_cond_exec): Likewise.
6327         (clrsbsi2): Likewise.
6328         (norm_f): Likewise.
6329         (normw): Likewise.
6330         (swap): Likewise.
6331         (divaw): Likewise.
6332         (flag): Likewise.
6333         (sr): Likewise.
6334         (kflag): Likewise.
6335         (ffs): Likewise.
6336         (ffs_f): Likewise.
6337         (fls): Likewise.
6338         (call_i): Remove 'S' asm letter, add jli instruction.
6339         (call_value_i): Likewise.
6340         * config/arc/arc.op (mjli-always): New option.
6341         * config/arc/constraints.md (Cji): New constraint.
6342         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
6343         operand.
6344         (subsf3_fpx): Likewise.
6345         (mulsf3_fpx): Likewise.
6346         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
6347         asm operand.
6348         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
6349         function attrbutes.
6350         * doc/invoke.texi (ARC): Document mjli-always option.
6351
6352 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
6353
6354         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
6355         avoid addition with 0 and use incw and decw where possible.
6356
6357 2018-01-26  Richard Biener  <rguenther@suse.de>
6358
6359         PR tree-optimization/81082
6360         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
6361         association if it requires casting to unsigned.
6362         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
6363         from fold_plusminus_mult_expr to catch important cases late when
6364         range info is available.
6365
6366 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6367
6368         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
6369         * configure.ac (hidden_linkonce): New test.
6370         * configure: Regenerate.
6371         * config.in: Regenerate.
6372
6373 2018-01-26  Julia Koval  <julia.koval@intel.com>
6374
6375         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
6376         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
6377         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
6378         _mm_mask_bitshuffle_epi64_mask): Fix type.
6379         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
6380         USI_FTYPE_V4DI_V4DI_USI): Remove.
6381         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
6382         __builtin_ia32_vpshufbitqmb256_mask,
6383         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
6384         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
6385         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
6386
6387 2018-01-26  Alan Modra  <amodra@gmail.com>
6388
6389         PR target/84033
6390         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
6391         UNSPEC_VBPERMQ.  Sort other unspecs.
6392
6393 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
6394
6395         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
6396
6397 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
6398
6399         PR middle-end/83055
6400         * predict.c (drop_profile): Do not push/pop cfun; update also
6401         node->count.
6402         (handle_missing_profiles): Fix logic looking for zero profiles.
6403
6404 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
6405
6406         PR middle-end/83977
6407         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
6408         on functions with #pragma omp declare simd or functions with simd
6409         attribute.
6410         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
6411         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
6412         Remove trailing \n from warning_at calls.
6413
6414 2018-01-25  Tom de Vries  <tom@codesourcery.com>
6415
6416         PR target/84028
6417         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6418         for neutered workers.
6419
6420 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
6421
6422         PR target/68467
6423         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
6424         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
6425
6426 2018-01-24  Jeff Law  <law@redhat.com>
6427
6428         PR target/83994
6429         * i386.c (get_probe_interval): Move to earlier point.
6430         (ix86_compute_frame_layout): If -fstack-clash-protection and
6431         the frame is larger than the probe interval, then use pushes
6432         to save registers rather than reg->mem moves.
6433         (ix86_expand_prologue): Remove conditional for int_registers_saved
6434         assertion.
6435
6436 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
6437
6438         PR target/84014
6439         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
6440         min/max for never referenced object.
6441
6442 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
6443
6444         PR middle-end/83977
6445         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
6446         here.
6447         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
6448         attributes from DECL_ATTRIBUTES (decl) without affecting
6449         DECL_ATTRIBUTES (current_function_decl).
6450         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
6451         functions with non-NULL DECL_ABSTRACT_ORIGIN.
6452
6453 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
6454
6455         PR tree-optimization/83979
6456         * fold-const.c (fold_comparison): Use constant_boolean_node
6457         instead of boolean_{true,false}_node.
6458
6459 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
6460
6461         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
6462         with zero counts.
6463
6464 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6465
6466         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6467         Simplify the clause that sets the length attribute.
6468         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6469         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
6470         clause that sets the length attribute.
6471         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6472
6473 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6474
6475         PR target/83589
6476         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
6477         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
6478         Add strict parameter.
6479         (prevent_branch_around_nothing): Insert dummy insn between branch to
6480         label and label with no ptx insn inbetween.
6481         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
6482
6483 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6484
6485         PR target/81352
6486         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6487         for neutered threads in warp.
6488         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
6489
6490 2018-01-24  Richard Biener  <rguenther@suse.de>
6491
6492         PR tree-optimization/83176
6493         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
6494         operands.
6495
6496 2018-01-24  Richard Biener  <rguenther@suse.de>
6497
6498         PR tree-optimization/82819
6499         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
6500         code generating pluses that are no-ops in the target precision.
6501
6502 2018-01-24  Richard Biener  <rguenther@suse.de>
6503
6504         PR middle-end/84000
6505         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
6506
6507 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6508
6509         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
6510         to merge probabilities.
6511         * predict.c (probably_never_executed): Also mark as cold functions
6512         with global 0 profile and guessed local profile.
6513         * profile-count.c (profile_probability::combine_with_count): New
6514         member function.
6515         * profile-count.h (profile_probability::operator*,
6516         profile_probability::operator*=, profile_probability::operator/,
6517         profile_probability::operator/=): Reduce precision to adjusted
6518         and set value to guessed on contradictory divisions.
6519         (profile_probability::combine_with_freq): Remove.
6520         (profile_probability::combine_wiht_count): Declare.
6521         (profile_count::force_nonzero):: Set to adjusted.
6522         (profile_count::probability_in):: Set quality to adjusted.
6523         * tree-ssa-tail-merge.c (replace_block_by): Use
6524         combine_with_count.
6525
6526 2018-01-23  Andrew Waterman  <andrew@sifive.com>
6527             Jim Wilson  <jimw@sifive.com>
6528
6529         * config/riscv/riscv.c (riscv_stack_boundary): New.
6530         (riscv_option_override): Set riscv_stack_boundary.  Handle
6531         riscv_preferred_stack_boundary_arg.
6532         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
6533         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
6534         (STACK_BOUNDARY): Set to riscv_stack_boundary.
6535         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
6536         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
6537         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
6538
6539 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
6540
6541         PR target/83905
6542         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
6543         of struct ix86_frame.
6544         (ix86_expand_epilogue): Likewise.  Add a local variable for
6545         the reg_save_offset field in struct ix86_frame.
6546
6547 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
6548
6549         PR tree-optimization/82604
6550         * tree-loop-distribution.c (enum partition_kind): New enum item
6551         PKIND_PARTIAL_MEMSET.
6552         (partition_builtin_p): Support above new enum item.
6553         (generate_code_for_partition): Ditto.
6554         (compute_access_range): Differentiate cases that equality can be
6555         proven at all loops, the innermost loops or no loops.
6556         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
6557         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
6558         (finalize_partitions, distribute_loop): Don't fuse partition of
6559         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
6560         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
6561         parloop is enabled.
6562
6563 2018-01-23  Martin Liska  <mliska@suse.cz>
6564
6565         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
6566         order to ignore the predictor.
6567         (PRED_POLYMORPHIC_CALL): Likewise.
6568         (PRED_RECURSIVE_CALL): Likewise.
6569
6570 2018-01-23  Martin Liska  <mliska@suse.cz>
6571
6572         * tree-profile.c (tree_profiling): Print function header to
6573         aware reader which function we are working on.
6574         * value-prof.c (gimple_find_values_to_profile): Do not print
6575         not interesting value histograms.
6576
6577 2018-01-23  Martin Liska  <mliska@suse.cz>
6578
6579         * profile-count.h (enum profile_quality): Add
6580         profile_uninitialized as the first value. Do not number values
6581         as they are zero based.
6582         (profile_count::verify): Update sanity check.
6583         (profile_probability::verify): Likewise.
6584
6585 2018-01-23  Nathan Sidwell  <nathan@acm.org>
6586
6587         * doc/invoke.texi (ffor-scope): Deprecate.
6588
6589 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6590
6591         PR tree-optimization/83510
6592         * domwalk.c (set_all_edges_as_executable): New function.
6593         (dom_walker::dom_walker): Convert bool param
6594         "skip_unreachable_blocks" to enum reachability.  Move setup of
6595         edge flags to set_all_edges_as_executable and only do it when
6596         reachability is REACHABLE_BLOCKS.
6597         * domwalk.h (enum dom_walker::reachability): New enum.
6598         (dom_walker::dom_walker): Convert bool param
6599         "skip_unreachable_blocks" to enum reachability.
6600         (set_all_edges_as_executable): New decl.
6601         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
6602         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
6603         "reachability".
6604         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
6605         but converting true to REACHABLE_BLOCKS.
6606         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
6607         * tree-vrp.c
6608         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
6609         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
6610         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
6611         REACHABLE_BLOCKS.
6612         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
6613         if check_all_array_refs will be called.
6614
6615 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6616
6617         * tree.c (selftest::test_location_wrappers): Add more test
6618         coverage.
6619
6620 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6621
6622         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
6623         (selftest::test_bit_in_range): Likewise.
6624
6625 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6626
6627         PR testsuite/83888
6628         * doc/sourcebuild.texi (vect_float): Say that the selector
6629         only describes the situation when -funsafe-math-optimizations is on.
6630         (vect_float_strict): Document.
6631
6632 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6633
6634         PR tree-optimization/83965
6635         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
6636         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
6637         instead of checking only for a reduction.
6638         (vect_recog_widen_sum_pattern): Likewise.
6639
6640 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6641
6642         * predict.c (probably_never_executed): Only use precise profile info.
6643         (compute_function_frequency): Skip after inlining hack since we now
6644         have quality checking.
6645
6646 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6647
6648         * profile-count.h (profile_probability::very_unlikely,
6649         profile_probability::unlikely, profile_probability::even): Set
6650         precision to guessed.
6651
6652 2018-01-23  Richard Biener  <rguenther@suse.de>
6653
6654         PR tree-optimization/83963
6655         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
6656         Properly terminate dominator walk when crossing the exit edge not
6657         when visiting its source block.
6658
6659 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
6660
6661         PR c++/83918
6662         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
6663         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
6664
6665 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
6666
6667         PR tree-optimization/83957
6668         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
6669         semicolon after for body surrounded by braces.
6670
6671         PR tree-optimization/83081
6672         * profile-count.h (profile_probability::split): New method.
6673         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
6674         Use profile_probability::split.
6675         (do_compare_rtx_and_jump): Fix adjustment of probabilities
6676         when splitting a single conditional jump into 2.
6677
6678 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
6679
6680         PR tree-optimization/69452
6681         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
6682         decl.
6683
6684 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6685
6686         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
6687         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
6688         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
6689
6690 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6691
6692         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
6693         * config/rl78/rl78.md (movdi): New define_expand.
6694         * config/rl78/rl78.c (rl78_split_movdi): New function.
6695
6696 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
6697
6698         PR target/83862
6699         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
6700         no longer used.
6701         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
6702         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
6703         128-bit to produce an UNSPEC move to get the double word with the
6704         signbit and then a shift directly to do signbit.
6705         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
6706         implementation with a new version that just does either a direct
6707         move or a regular move.  Move memory interface to separate insns.
6708         Move insns so they are next to the expander.
6709         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
6710         with signbit move.  Split big and little endian case.
6711         (signbit<mode>2_dm_mem_le): Likewise.
6712         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
6713         (signbit<mode>2_dm2): Likewise.
6714
6715 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6716
6717         * config/rl78/rl78.md (anddi3): New define_expand.
6718
6719 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6720
6721         * config/rl78/rl78.md (umindi3): New define_expand.
6722
6723 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6724
6725         * config/rl78/rl78.md (smindi3): New define_expand.
6726
6727 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6728
6729         * config/rl78/rl78.md (smaxdi3): New define_expand.
6730
6731 2018-01-22  Carl Love  <cel@us.ibm.com>
6732
6733         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
6734         LVX_V1TI): Add macro expansion.
6735         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
6736         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
6737         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
6738         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
6739         Change check to determine if the instruction is a byte reversing
6740         entry.  Fix typo in comment.
6741         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
6742         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
6743         Add def_builtin calls for new builtins.
6744         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
6745         Add define_insn expansion.
6746
6747 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6748
6749         * config/rl78/rl78.md (umaxdi3): New define_expand.
6750
6751 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6752
6753         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
6754         for non-QImode registers.
6755
6756 2018-01-22  Richard Biener  <rguenther@suse.de>
6757
6758         PR tree-optimization/83963
6759         * graphite-scop-detection.c (scop_detection::get_sese): Delay
6760         including the loop exit block.
6761         (scop_detection::merge_sese): Likewise.
6762         (scop_detection::add_scop): Do it here instead.
6763
6764 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6765
6766         * doc/sourcebuild.texi (arm_softfloat): Document.
6767
6768 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
6769
6770         PR gcc/77734
6771         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
6772         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
6773         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
6774
6775 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6776             David Edelsohn  <dje.gcc@gmail.com>
6777
6778         PR target/83946
6779         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6780         Change "crset eq" to "crset 2".
6781         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6782         (*call_indirect_aix<mode>_nospec): Likewise.
6783         (*call_value_indirect_aix<mode>_nospec): Likewise.
6784         (*call_indirect_elfv2<mode>_nospec): Likewise.
6785         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
6786         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
6787         change assembly output from . to $.
6788         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6789         (indirect_jump<mode>_nospec): Change assembly output from . to $.
6790         (*tablejump<mode>_internal1_nospec): Likewise.
6791
6792 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
6793
6794         PR target/80870
6795         * config/sh/sh_optimize_sett_clrt.cc:
6796         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
6797
6798 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
6799
6800         PR tree-optimization/83940
6801         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
6802         offset_dt to vect_constant_def rather than vect_unknown_def_type.
6803         (vect_check_load_store_mask): Add a mask_dt_out parameter and
6804         use it to pass back the definition type.
6805         (vect_check_store_rhs): Likewise rhs_dt_out.
6806         (vect_build_gather_load_calls): Add a mask_dt argument and use
6807         it instead of a call to vect_is_simple_use.
6808         (vectorizable_store): Update calls to vect_check_load_store_mask
6809         and vect_check_store_rhs.  Use the dt returned by the latter instead
6810         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
6811         instead of calls to vect_is_simple_use.  Pass the scalar rather
6812         than the vector operand to vect_is_simple_use when handling
6813         second and subsequent copies of an rhs value.
6814         (vectorizable_load): Update calls to vect_check_load_store_mask
6815         and vect_build_gather_load_calls.  Use the cached mask_dt and
6816         gs_info.offset_dt instead of calls to vect_is_simple_use.
6817
6818 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
6819
6820         PR middle-end/83945
6821         * tree-emutls.c: Include gimplify.h.
6822         (lower_emutls_2): New function.
6823         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
6824         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
6825         it before further processing.
6826
6827         PR target/83930
6828         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
6829         UINTVAL (trueop1) instead of INTVAL (op1).
6830
6831 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
6832
6833         PR debug/81570
6834         PR debug/83728
6835         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
6836         INCOMING_FRAME_SP_OFFSET if not defined.
6837         (scan_trace): Add ENTRY argument.  If true and
6838         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
6839         emit a note to adjust the CFA offset.
6840         (create_cfi_notes): Adjust scan_trace callers.
6841         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
6842         INCOMING_FRAME_SP_OFFSET in the CIE.
6843         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
6844         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
6845         Likewise.
6846         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
6847         * doc/tm.texi: Regenerated.
6848
6849 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6850
6851         PR rtl-optimization/83147
6852         * lra-constraints.c (remove_inheritance_pseudos): Use
6853         lra_substitute_pseudo_within_insn.
6854
6855 2018-01-19  Tom de Vries  <tom@codesourcery.com>
6856             Cesar Philippidis  <cesar@codesourcery.com>
6857
6858         PR target/83920
6859         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
6860
6861 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
6862
6863         PR target/83790
6864         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
6865         spaces for function labels.
6866
6867 2018-01-19  Martin Liska  <mliska@suse.cz>
6868
6869         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
6870         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
6871         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
6872         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
6873         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
6874         (PRED_CONST_RETURN): Change from 69 to 65.
6875         (PRED_NULL_RETURN): Change from 91 to 71.
6876         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
6877         (PRED_LOOP_GUARD): Change from 66 to 73.
6878
6879 2018-01-19  Martin Liska  <mliska@suse.cz>
6880
6881         * predict.c (predict_insn_def): Add new assert.
6882         (struct branch_predictor): Change type to signed integer.
6883         (test_prediction_value_range): Amend test to cover
6884         PROB_UNINITIALIZED.
6885         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
6886         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
6887         (PRED_LOOP_ITERATIONS_MAX): Likewise.
6888         (PRED_LOOP_IV_COMPARE): Likewise.
6889         * predict.h (PROB_UNINITIALIZED): Define new constant.
6890
6891 2018-01-19  Martin Liska  <mliska@suse.cz>
6892
6893         * predict.c (dump_prediction): Add new format for
6894         analyze_brprob.py script which is enabled with -details
6895         suboption.
6896         * profile-count.h (precise_p): New function.
6897
6898 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6899
6900         PR tree-optimization/83922
6901         * tree-vect-loop.c (vect_verify_full_masking): Return false if
6902         there are no statements that need masking.
6903         (vect_active_double_reduction_p): New function.
6904         (vect_analyze_loop_operations): Use it when handling phis that
6905         are not in the loop header.
6906
6907 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6908
6909         PR tree-optimization/83914
6910         * tree-vect-loop.c (vectorizable_induction): Don't convert
6911         init_expr or apply the peeling adjustment for inductions
6912         that are nested within the vectorized loop.
6913
6914 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6915
6916         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
6917         instead of NEG.
6918
6919 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
6920
6921         PR sanitizer/81715
6922         PR testsuite/83882
6923         * function.h (gimplify_parameters): Add gimple_seq * argument.
6924         * function.c: Include gimple.h and options.h.
6925         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
6926         for the added local temporaries if needed.
6927         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
6928         if there are any parameter cleanups, wrap whole body into a
6929         try/finally with the cleanups.
6930
6931 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
6932
6933         PR target/82964
6934         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
6935         Use GET_MODE_CLASS for scalar floating point.
6936
6937 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6938
6939         PR ipa/82256
6940         patch by PaX Team
6941         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
6942         Fix call of call_cgraph_insertion_hooks.
6943
6944 2018-01-18  Martin Sebor  <msebor@redhat.com>
6945
6946         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
6947
6948 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6949
6950         PR ipa/83619
6951         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
6952         frequencies.
6953
6954 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
6955
6956         PR other/70268
6957         * common.opt: (-ffile-prefix-map): New option.
6958         * opts.c (common_handle_option): Defer it.
6959         * opts-global.c (handle_common_deferred_options): Handle it.
6960         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
6961         * file-prefix-map.h: New file.
6962         (remap_debug_filename, add_debug_prefix_map): ...here.
6963         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
6964         * final.c (debug_prefix_map, add_debug_prefix_map
6965         remap_debug_filename): Move to...
6966         * file-prefix-map.c: New file.
6967         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
6968         generalize, get rid of alloca(), use strrchr() instead of strchr().
6969         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
6970         Implement in terms of add_prefix_map().
6971         (remap_macro_filename, remap_debug_filename): Implement in term of
6972         remap_filename().
6973         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
6974         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
6975         * dbxout.c: Include file-prefix-map.h.
6976         * varasm.c: Likewise.
6977         * vmsdbgout.c: Likewise.
6978         * xcoffout.c: Likewise.
6979         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
6980         * doc/cppopts.texi (-fmacro-prefix-map): Document.
6981         * doc/invoke.texi (-ffile-prefix-map): Document.
6982         (-fdebug-prefix-map): Update description.
6983
6984 2018-01-18  Martin Liska  <mliska@suse.cz>
6985
6986         * config/i386/i386.c (indirect_thunk_name): Document that also
6987         lfence is emitted.
6988         (output_indirect_thunk): Document why both instructions
6989         (pause and lfence) are generated.
6990
6991 2018-01-18  Richard Biener  <rguenther@suse.de>
6992
6993         PR tree-optimization/83887
6994         * graphite-scop-detection.c
6995         (scop_detection::get_nearest_dom_with_single_entry): Remove.
6996         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
6997         (scop_detection::merge_sese): Re-implement with a flood-fill
6998         algorithm that properly finds a SESE region if it exists.
6999
7000 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
7001
7002         PR c/61240
7003         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
7004         pointer_diff optimizations use view_convert instead of convert.
7005
7006 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7007
7008         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
7009         Generate different code for -mno-speculate-indirect-jumps.
7010         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
7011         (*call_indirect_aix<mode>): Disable for
7012         -mno-speculate-indirect-jumps.
7013         (*call_indirect_aix<mode>_nospec): New define_insn.
7014         (*call_value_indirect_aix<mode>): Disable for
7015         -mno-speculate-indirect-jumps.
7016         (*call_value_indirect_aix<mode>_nospec): New define_insn.
7017         (*sibcall_nonlocal_sysv<mode>): Generate different code for
7018         -mno-speculate-indirect-jumps.
7019         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
7020
7021 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
7022
7023         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
7024         long double type, set the flags for noting the default long double
7025         type, even if we don't pass or return a long double type.
7026
7027 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
7028
7029         PR ipa/83051
7030         * ipa-inline.c (flatten_function): Do not overwrite final inlining
7031         failure.
7032
7033 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
7034
7035         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
7036         support for merge[hl].
7037         (fold_mergehl_helper): New helper function.
7038         (tree-vector-builder.h): New #include for tree_vector_builder usage.
7039         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
7040         (altivec_vmrglw_direct): Add xxmrglw insn.
7041
7042 2018-01-17  Andrew Waterman  <andrew@sifive.com>
7043
7044         * config/riscv/riscv.c (riscv_conditional_register_usage): If
7045         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
7046
7047 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
7048
7049         PR lto/83121
7050         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
7051         call the lto_location_cache before reading the
7052         DECL_SOURCE_LOCATION of the types.
7053
7054 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
7055             Richard Sandiford  <richard.sandiford@linaro.org>
7056
7057         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
7058         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
7059         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
7060         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
7061         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
7062         Add declaration.
7063         * config/aarch64/constraints.md (aarch64_movti_operand):
7064         Limit immediates.
7065         * config/aarch64/predicates.md (Uti): Add new constraint.
7066
7067 2018-01-17 Carl Love  <cel@us.ibm.com>
7068
7069         * config/rs6000/vsx.md (define_expand xl_len_r,
7070         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
7071         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
7072         lxvll.
7073         (define_expand, define_insn): Move the shift left from  the
7074         define_insn to the define_expand for lxvl and stxvl instructions.
7075         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
7076         and XL_LEN_R definitions to PURE.
7077
7078 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
7079
7080         * config/i386/i386.c (indirect_thunk_name): Declare regno
7081         as unsigned int.  Compare regno with INVALID_REGNUM.
7082         (output_indirect_thunk): Ditto.
7083         (output_indirect_thunk_function): Ditto.
7084         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
7085         in the call to output_indirect_thunk_function.
7086
7087 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
7088
7089         PR middle-end/83884
7090         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
7091         rather than the size of inner_type to determine the stack slot size
7092         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
7093
7094 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
7095
7096         PR target/83546
7097         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
7098         to PTA_SILVERMONT.
7099
7100 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
7101
7102         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
7103         endian Linux systems to optionally enable multilibs for selecting
7104         the long double type if the user configured an explicit type.
7105         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
7106         have no long double multilibs if not defined.
7107         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
7108         warn if the user used -mabi={ieee,ibm}longdouble and we built
7109         multilibs for long double.
7110         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
7111         appropriate multilib option.
7112         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
7113         multilib options.
7114         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
7115         for building long double multilibs.
7116         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
7117
7118 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
7119
7120         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
7121         copies.
7122
7123         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
7124         64 bits.
7125         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
7126         128 bits.
7127
7128         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
7129         variables.
7130
7131         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
7132         return value.
7133
7134 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
7135
7136         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
7137         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
7138
7139 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7140
7141         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
7142         different rtl trees depending on TARGET_64BIT.
7143         (rs6000_gen_lvx): Likewise.
7144
7145 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
7146
7147         * config/visium/visium.md (nop): Tweak comment.
7148         (hazard_nop): Likewise.
7149
7150 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7151
7152         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
7153         -mspeculate-indirect-jumps.
7154         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
7155         for -mno-speculate-indirect-jumps.
7156         (*call_indirect_elfv2<mode>_nospec): New define_insn.
7157         (*call_value_indirect_elfv2<mode>): Disable for
7158         -mno-speculate-indirect-jumps.
7159         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
7160         (indirect_jump): Emit different RTL for
7161         -mno-speculate-indirect-jumps.
7162         (*indirect_jump<mode>): Disable for
7163         -mno-speculate-indirect-jumps.
7164         (*indirect_jump<mode>_nospec): New define_insn.
7165         (tablejump): Emit different RTL for
7166         -mno-speculate-indirect-jumps.
7167         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
7168         (tablejumpsi_nospec): New define_expand.
7169         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
7170         (tablejumpdi_nospec): New define_expand.
7171         (*tablejump<mode>_internal1): Disable for
7172         -mno-speculate-indirect-jumps.
7173         (*tablejump<mode>_internal1_nospec): New define_insn.
7174         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
7175         option.
7176
7177 2018-01-16  Artyom Skrobov tyomitch@gmail.com
7178
7179         * caller-save.c (insert_save): Drop unnecessary parameter.  All
7180         callers updated.
7181
7182 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7183             Richard Biener  <rguenth@suse.de>
7184
7185         PR libgomp/83590
7186         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
7187         return early, inline manually is_gimple_sizepos.  Make sure if we
7188         call gimplify_expr we don't end up with a gimple constant.
7189         * tree.c (variably_modified_type_p): Don't return true for
7190         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
7191         * gimplify.h (is_gimple_sizepos): Remove.
7192
7193 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7194
7195         PR tree-optimization/83857
7196         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
7197         vectorizable_live_operation for pure SLP statements.
7198         (vectorizable_live_operation): Handle PHIs.
7199
7200 2018-01-16  Richard Biener  <rguenther@suse.de>
7201
7202         PR tree-optimization/83867
7203         * tree-vect-stmts.c (vect_transform_stmt): Precompute
7204         nested_in_vect_loop_p since the scalar stmt may get invalidated.
7205
7206 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7207
7208         PR c/83844
7209         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
7210         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
7211         If off is not INTEGER_CST, issue a may not be aligned warning
7212         rather than isn't aligned.  Use isn%'t rather than isn't.
7213         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
7214         into MULT_EXPR.
7215         <case MULT_EXPR>: Improve the case when bottom and one of the
7216         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
7217         operand, in that case check if the other operand is multiple of
7218         bottom divided by the INTEGER_CST operand.
7219
7220 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7221
7222         PR target/83858
7223         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
7224         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
7225         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
7226         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
7227         * config/pa/pa.c (pa_function_arg_advance): Likewise.
7228         (pa_function_arg, pa_arg_partial_bytes): Likewise.
7229         (pa_function_arg_size): New function.
7230
7231 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7232
7233         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
7234         in a separate statement.
7235
7236 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7237
7238         PR tree-optimization/83847
7239         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
7240         group gathers and scatters.
7241
7242 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7243
7244         PR rtl-optimization/86620
7245         * params.def (max-sched-ready-insns): Bump minimum value to 1.
7246
7247         PR rtl-optimization/83213
7248         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
7249         to last if both are JUMP_INSNs.
7250
7251         PR tree-optimization/83843
7252         * gimple-ssa-store-merging.c
7253         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
7254         store_immediate_info for bswap/nop orig_stores.
7255
7256 2018-01-15  Andrew Waterman  <andrew@sifive.com>
7257
7258         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
7259         !TARGET_MUL.
7260         <UDIV>: Increase cost if !TARGET_DIV.
7261
7262 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
7263
7264         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
7265         (define_attr "cr_logical_3op"): New.
7266         (cceq_ior_compare): Adjust.
7267         (cceq_ior_compare_complement): Adjust.
7268         (*cceq_rev_compare): Adjust.
7269         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
7270         (is_cracked_insn): Adjust.
7271         (insn_must_be_first_in_group): Adjust.
7272         * config/rs6000/40x.md: Adjust.
7273         * config/rs6000/440.md: Adjust.
7274         * config/rs6000/476.md: Adjust.
7275         * config/rs6000/601.md: Adjust.
7276         * config/rs6000/603.md: Adjust.
7277         * config/rs6000/6xx.md: Adjust.
7278         * config/rs6000/7450.md: Adjust.
7279         * config/rs6000/7xx.md: Adjust.
7280         * config/rs6000/8540.md: Adjust.
7281         * config/rs6000/cell.md: Adjust.
7282         * config/rs6000/e300c2c3.md: Adjust.
7283         * config/rs6000/e500mc.md: Adjust.
7284         * config/rs6000/e500mc64.md: Adjust.
7285         * config/rs6000/e5500.md: Adjust.
7286         * config/rs6000/e6500.md: Adjust.
7287         * config/rs6000/mpc.md: Adjust.
7288         * config/rs6000/power4.md: Adjust.
7289         * config/rs6000/power5.md: Adjust.
7290         * config/rs6000/power6.md: Adjust.
7291         * config/rs6000/power7.md: Adjust.
7292         * config/rs6000/power8.md: Adjust.
7293         * config/rs6000/power9.md: Adjust.
7294         * config/rs6000/rs64.md: Adjust.
7295         * config/rs6000/titan.md: Adjust.
7296
7297 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7298
7299         * config/i386/predicates.md (indirect_branch_operand): Rewrite
7300         ix86_indirect_branch_register logic.
7301
7302 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7303
7304         * config/i386/constraints.md (Bs): Update
7305         ix86_indirect_branch_register check.  Don't check
7306         ix86_indirect_branch_register with GOT_memory_operand.
7307         (Bw): Likewise.
7308         * config/i386/predicates.md (GOT_memory_operand): Don't check
7309         ix86_indirect_branch_register here.
7310         (GOT32_symbol_operand): Likewise.
7311
7312 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7313
7314         * config/i386/predicates.md (constant_call_address_operand):
7315         Rewrite ix86_indirect_branch_register logic.
7316         (sibcall_insn_operand): Likewise.
7317
7318 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7319
7320         * config/i386/constraints.md (Bs): Replace
7321         ix86_indirect_branch_thunk_register with
7322         ix86_indirect_branch_register.
7323         (Bw): Likewise.
7324         * config/i386/i386.md (indirect_jump): Likewise.
7325         (tablejump): Likewise.
7326         (*sibcall_memory): Likewise.
7327         (*sibcall_value_memory): Likewise.
7328         Peepholes of indirect call and jump via memory: Likewise.
7329         * config/i386/i386.opt: Likewise.
7330         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7331         (GOT_memory_operand): Likewise.
7332         (call_insn_operand): Likewise.
7333         (sibcall_insn_operand): Likewise.
7334         (GOT32_symbol_operand): Likewise.
7335
7336 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
7337
7338         PR middle-end/83837
7339         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
7340         type rather than type addr's type points to.
7341         (expand_omp_atomic_mutex): Likewise.
7342         (expand_omp_atomic): Likewise.
7343
7344 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7345
7346         PR target/83839
7347         * config/i386/i386.c (output_indirect_thunk_function): Use
7348         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
7349         for  __x86_return_thunk.
7350
7351 2018-01-15  Richard Biener  <rguenther@suse.de>
7352
7353         PR middle-end/83850
7354         * expmed.c (extract_bit_field_1): Fix typo.
7355
7356 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7357
7358         PR target/83687
7359         * config/arm/iterators.md (VF): New mode iterator.
7360         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
7361         Remove integer-related logic from pattern.
7362         (neon_vabd<mode>_3): Likewise.
7363
7364 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
7365
7366         PR middle-end/82694
7367         * common.opt (fstrict-overflow): No longer an alias.
7368         (fwrapv-pointer): New option.
7369         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
7370         also for pointer types based on flag_wrapv_pointer.
7371         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
7372         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
7373         opts->x_flag_wrapv got set.
7374         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
7375         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
7376         POINTER_TYPE_OVERFLOW_UNDEFINED.
7377         * match.pd: Likewise in address comparison pattern.
7378         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
7379
7380 2018-01-15  Richard Biener  <rguenther@suse.de>
7381
7382         PR lto/83804
7383         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
7384         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
7385         Reset type names to their identifier if their TYPE_DECL doesn't
7386         have linkage (and thus is used for ODR and devirt).
7387         (save_debug_info_for_decl): Remove.
7388         (save_debug_info_for_type): Likewise.
7389         (add_tree_to_fld_list): Adjust.
7390         * tree-pretty-print.c (dump_generic_node): Make dumping of
7391         type names more robust.
7392
7393 2018-01-15  Richard Biener  <rguenther@suse.de>
7394
7395         * BASE-VER: Bump to 8.0.1.
7396
7397 2018-01-14  Martin Sebor  <msebor@redhat.com>
7398
7399         PR other/83508
7400         * builtins.c (check_access): Avoid warning when the no-warning bit
7401         is set.
7402
7403 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
7404
7405         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
7406         * ira-color (allocno_hard_regs_compare): Likewise.
7407
7408 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
7409
7410         PR target/83013
7411         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
7412         Use .pushsection/.popsection.
7413
7414 2018-01-14  Martin Sebor  <msebor@redhat.com>
7415
7416         PR c++/81327
7417         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
7418
7419 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
7420
7421         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
7422         entry from extra_headers.
7423         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
7424         extra_headers, make the list bitwise identical to the i?86-*-* one.
7425
7426 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7427
7428         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
7429         -mcmodel=large with -mindirect-branch=thunk,
7430         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
7431         -mfunction-return=thunk-extern.
7432         * doc/invoke.texi: Document -mcmodel=large is incompatible with
7433         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
7434         -mfunction-return=thunk and -mfunction-return=thunk-extern.
7435
7436 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7437
7438         * config/i386/i386.c (print_reg): Print the name of the full
7439         integer register without '%'.
7440         (ix86_print_operand): Handle 'V'.
7441          * doc/extend.texi: Document 'V' modifier.
7442
7443 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7444
7445         * config/i386/constraints.md (Bs): Disallow memory operand for
7446         -mindirect-branch-register.
7447         (Bw): Likewise.
7448         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7449         (GOT_memory_operand): Likewise.
7450         (call_insn_operand): Likewise.
7451         (sibcall_insn_operand): Likewise.
7452         (GOT32_symbol_operand): Likewise.
7453         * config/i386/i386.md (indirect_jump): Call convert_memory_address
7454         for -mindirect-branch-register.
7455         (tablejump): Likewise.
7456         (*sibcall_memory): Likewise.
7457         (*sibcall_value_memory): Likewise.
7458         Disallow peepholes of indirect call and jump via memory for
7459         -mindirect-branch-register.
7460         (*call_pop): Replace m with Bw.
7461         (*call_value_pop): Likewise.
7462         (*sibcall_pop_memory): Replace m with Bs.
7463         * config/i386/i386.opt (mindirect-branch-register): New option.
7464         * doc/invoke.texi: Document -mindirect-branch-register option.
7465
7466 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7467
7468         * config/i386/i386-protos.h (ix86_output_function_return): New.
7469         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
7470         set function_return_type.
7471         (indirect_thunk_name): Add ret_p to indicate thunk for function
7472         return.
7473         (output_indirect_thunk_function): Pass false to
7474         indirect_thunk_name.
7475         (ix86_output_indirect_branch_via_reg): Likewise.
7476         (ix86_output_indirect_branch_via_push): Likewise.
7477         (output_indirect_thunk_function): Create alias for function
7478         return thunk if regno < 0.
7479         (ix86_output_function_return): New function.
7480         (ix86_handle_fndecl_attribute): Handle function_return.
7481         (ix86_attribute_table): Add function_return.
7482         * config/i386/i386.h (machine_function): Add
7483         function_return_type.
7484         * config/i386/i386.md (simple_return_internal): Use
7485         ix86_output_function_return.
7486         (simple_return_internal_long): Likewise.
7487         * config/i386/i386.opt (mfunction-return=): New option.
7488         (indirect_branch): Mention -mfunction-return=.
7489         * doc/extend.texi: Document function_return function attribute.
7490         * doc/invoke.texi: Document -mfunction-return= option.
7491
7492 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7493
7494         * config/i386/i386-opts.h (indirect_branch): New.
7495         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
7496         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
7497         with local indirect jump when converting indirect call and jump.
7498         (ix86_set_indirect_branch_type): New.
7499         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
7500         (indirectlabelno): New.
7501         (indirect_thunk_needed): Likewise.
7502         (indirect_thunk_bnd_needed): Likewise.
7503         (indirect_thunks_used): Likewise.
7504         (indirect_thunks_bnd_used): Likewise.
7505         (INDIRECT_LABEL): Likewise.
7506         (indirect_thunk_name): Likewise.
7507         (output_indirect_thunk): Likewise.
7508         (output_indirect_thunk_function): Likewise.
7509         (ix86_output_indirect_branch_via_reg): Likewise.
7510         (ix86_output_indirect_branch_via_push): Likewise.
7511         (ix86_output_indirect_branch): Likewise.
7512         (ix86_output_indirect_jmp): Likewise.
7513         (ix86_code_end): Call output_indirect_thunk_function if needed.
7514         (ix86_output_call_insn): Call ix86_output_indirect_branch if
7515         needed.
7516         (ix86_handle_fndecl_attribute): Handle indirect_branch.
7517         (ix86_attribute_table): Add indirect_branch.
7518         * config/i386/i386.h (machine_function): Add indirect_branch_type
7519         and has_local_indirect_jump.
7520         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
7521         to true.
7522         (tablejump): Likewise.
7523         (*indirect_jump): Use ix86_output_indirect_jmp.
7524         (*tablejump_1): Likewise.
7525         (simple_return_indirect_internal): Likewise.
7526         * config/i386/i386.opt (mindirect-branch=): New option.
7527         (indirect_branch): New.
7528         (keep): Likewise.
7529         (thunk): Likewise.
7530         (thunk-inline): Likewise.
7531         (thunk-extern): Likewise.
7532         * doc/extend.texi: Document indirect_branch function attribute.
7533         * doc/invoke.texi: Document -mindirect-branch= option.
7534
7535 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
7536
7537         PR ipa/83051
7538         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
7539
7540 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
7541
7542         * ipa-inline.c (want_inline_small_function_p): Return false if
7543         inlining has already failed with CIF_FINAL_ERROR.
7544         (update_caller_keys): Call want_inline_small_function_p before
7545         can_inline_edge_p.
7546         (update_callee_keys): Likewise.
7547
7548 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7549
7550         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
7551         New function.
7552         (rs6000_quadword_masked_address_p): Likewise.
7553         (quad_aligned_load_p): Likewise.
7554         (quad_aligned_store_p): Likewise.
7555         (const_load_sequence_p): Add comment to describe the outer-most loop.
7556         (mimic_memory_attributes_and_flags): New function.
7557         (rs6000_gen_stvx): Likewise.
7558         (replace_swapped_aligned_store): Likewise.
7559         (rs6000_gen_lvx): Likewise.
7560         (replace_swapped_aligned_load): Likewise.
7561         (replace_swapped_load_constant): Capitalize argument name in
7562         comment describing this function.
7563         (rs6000_analyze_swaps): Add a third pass to search for vector loads
7564         and stores that access quad-word aligned addresses and replace
7565         with stvx or lvx instructions when appropriate.
7566         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
7567         New function prototype.
7568         (rs6000_quadword_masked_address_p): Likewise.
7569         (rs6000_gen_lvx): Likewise.
7570         (rs6000_gen_stvx): Likewise.
7571         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
7572         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
7573         when memory address is aligned.
7574         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
7575         this split to select lvx instruction when memory address is aligned.
7576         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
7577         instruction when memory address is aligned.
7578         (*vsx_le_perm_load_v16qi): Likewise.
7579         (four unnamed splitters): Modify to select the stvx instruction
7580         when memory is aligned.
7581
7582 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
7583
7584         * predict.c (determine_unlikely_bbs): Handle correctly BBs
7585         which appears in the queue multiple times.
7586
7587 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7588             Alan Hayward  <alan.hayward@arm.com>
7589             David Sherwood  <david.sherwood@arm.com>
7590
7591         * tree-vectorizer.h (vec_lower_bound): New structure.
7592         (_loop_vec_info): Add check_nonzero and lower_bounds.
7593         (LOOP_VINFO_CHECK_NONZERO): New macro.
7594         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
7595         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
7596         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
7597         fields.  Make seg_len the distance travelled, not including the
7598         access size.
7599         (dr_direction_indicator): Declare.
7600         (dr_zero_step_indicator): Likewise.
7601         (dr_known_forward_stride_p): Likewise.
7602         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
7603         tree-ssanames.h.
7604         (runtime_alias_check_p): Allow runtime alias checks with
7605         variable strides.
7606         (operator ==): Compare access_size and align.
7607         (prune_runtime_alias_test_list): Rework for new distinction between
7608         the access_size and seg_len.
7609         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
7610         segment lengths.
7611         (get_segment_min_max): New function.
7612         (create_intersect_range_checks): Use it.
7613         (dr_step_indicator): New function.
7614         (dr_direction_indicator): Likewise.
7615         (dr_zero_step_indicator): Likewise.
7616         (dr_known_forward_stride_p): Likewise.
7617         * tree-loop-distribution.c (data_ref_segment_size): Return
7618         DR_STEP * (niters - 1).
7619         (compute_alias_check_pairs): Update call to the dr_with_seg_len
7620         constructor.
7621         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
7622         (vect_preserves_scalar_order_p): New function, split out from...
7623         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
7624         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
7625         (vect_vfa_access_size): New function.
7626         (vect_vfa_align): Likewise.
7627         (vect_compile_time_alias): Take access_size_a and access_b arguments.
7628         (dump_lower_bound): New function.
7629         (vect_check_lower_bound): Likewise.
7630         (vect_small_gap_p): Likewise.
7631         (vectorizable_with_step_bound_p): Likewise.
7632         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
7633         depencies if the vectorization factor is 1.  Convert the checks
7634         for nonzero steps into checks on the bounds of DR_STEP.  Try using
7635         a bunds check for variable steps if the minimum required step is
7636         relatively small. Update calls to the dr_with_seg_len
7637         constructor and to vect_compile_time_alias.
7638         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
7639         function.
7640         (vect_loop_versioning): Call it.
7641         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
7642         when retrying.
7643         (vect_estimate_min_profitable_iters): Account for any bounds checks.
7644
7645 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7646             Alan Hayward  <alan.hayward@arm.com>
7647             David Sherwood  <david.sherwood@arm.com>
7648
7649         * doc/sourcebuild.texi (vect_scatter_store): Document.
7650         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
7651         optabs.
7652         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
7653         Document.
7654         * genopinit.c (main): Add supports_vec_scatter_store and
7655         supports_vec_scatter_store_cached to target_optabs.
7656         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
7657         IFN_MASK_SCATTER_STORE.
7658         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
7659         functions.
7660         * internal-fn.h (internal_store_fn_p): Declare.
7661         (internal_fn_stored_value_index): Likewise.
7662         * internal-fn.c (scatter_store_direct): New macro.
7663         (expand_scatter_store_optab_fn): New function.
7664         (direct_scatter_store_optab_supported_p): New macro.
7665         (internal_store_fn_p): New function.
7666         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
7667         IFN_MASK_SCATTER_STORE.
7668         (internal_fn_mask_index): Likewise.
7669         (internal_fn_stored_value_index): New function.
7670         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
7671         for scatter stores.
7672         * optabs-query.h (supports_vec_scatter_store_p): Declare.
7673         * optabs-query.c (supports_vec_scatter_store_p): New function.
7674         * tree-vectorizer.h (vect_get_store_rhs): Declare.
7675         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
7676         true for scatter stores.
7677         (vect_gather_scatter_fn_p): Handle scatter stores too.
7678         (vect_check_gather_scatter): Consider using scatter stores if
7679         supports_vec_scatter_store_p.
7680         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
7681         scatter stores too.
7682         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7683         internal_fn_stored_value_index.
7684         (check_load_store_masking): Handle scatter stores too.
7685         (vect_get_store_rhs): Make public.
7686         (vectorizable_call): Use internal_store_fn_p.
7687         (vectorizable_store): Handle scatter store internal functions.
7688         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
7689         when deciding whether the end of the group has been reached.
7690         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
7691         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
7692         (mask_scatter_store<mode>): New insns.
7693
7694 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7695             Alan Hayward  <alan.hayward@arm.com>
7696             David Sherwood  <david.sherwood@arm.com>
7697
7698         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
7699         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
7700         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
7701         function.
7702         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
7703         Use vect_truncate_gather_scatter_offset if we can't treat the
7704         operation as a normal gather load or scatter store.
7705         (get_group_load_store_type): Take the gather_scatter_info
7706         as argument.  Try using a gather load or scatter store for
7707         single-element groups.
7708         (get_load_store_type): Update calls to get_group_load_store_type
7709         and vect_use_strided_gather_scatters_p.
7710
7711 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7712             Alan Hayward  <alan.hayward@arm.com>
7713             David Sherwood  <david.sherwood@arm.com>
7714
7715         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
7716         optional tree argument.
7717         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
7718         null target hooks.
7719         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
7720         but continue to use the current value as a fallback.
7721         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
7722         to compare the updates.
7723         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
7724         (get_load_store_type): Use it when handling a strided access.
7725         (vect_get_strided_load_store_ops): New function.
7726         (vect_get_data_ptr_increment): Likewise.
7727         (vectorizable_load): Handle strided gather loads.  Always pass
7728         a step to vect_create_data_ref_ptr and bump_vector_ptr.
7729
7730 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7731             Alan Hayward  <alan.hayward@arm.com>
7732             David Sherwood  <david.sherwood@arm.com>
7733
7734         * doc/md.texi (gather_load@var{m}): Document.
7735         (mask_gather_load@var{m}): Likewise.
7736         * genopinit.c (main): Add supports_vec_gather_load and
7737         supports_vec_gather_load_cached to target_optabs.
7738         * optabs-tree.c (init_tree_optimization_optabs): Use
7739         ggc_cleared_alloc to allocate target_optabs.
7740         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
7741         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
7742         functions.
7743         * internal-fn.h (internal_load_fn_p): Declare.
7744         (internal_gather_scatter_fn_p): Likewise.
7745         (internal_fn_mask_index): Likewise.
7746         (internal_gather_scatter_fn_supported_p): Likewise.
7747         * internal-fn.c (gather_load_direct): New macro.
7748         (expand_gather_load_optab_fn): New function.
7749         (direct_gather_load_optab_supported_p): New macro.
7750         (direct_internal_fn_optab): New function.
7751         (internal_load_fn_p): Likewise.
7752         (internal_gather_scatter_fn_p): Likewise.
7753         (internal_fn_mask_index): Likewise.
7754         (internal_gather_scatter_fn_supported_p): Likewise.
7755         * optabs-query.c (supports_at_least_one_mode_p): New function.
7756         (supports_vec_gather_load_p): Likewise.
7757         * optabs-query.h (supports_vec_gather_load_p): Declare.
7758         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
7759         and memory_type field.
7760         (NUM_PATTERNS): Bump to 15.
7761         * tree-vect-data-refs.c: Include internal-fn.h.
7762         (vect_gather_scatter_fn_p): New function.
7763         (vect_describe_gather_scatter_call): Likewise.
7764         (vect_check_gather_scatter): Try using internal functions for
7765         gather loads.  Recognize existing calls to a gather load function.
7766         (vect_analyze_data_refs): Consider using gather loads if
7767         supports_vec_gather_load_p.
7768         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
7769         (vect_get_gather_scatter_offset_type): Likewise.
7770         (vect_convert_mask_for_vectype): Likewise.
7771         (vect_add_conversion_to_patterm): Likewise.
7772         (vect_try_gather_scatter_pattern): Likewise.
7773         (vect_recog_gather_scatter_pattern): New pattern recognizer.
7774         (vect_vect_recog_func_ptrs): Add it.
7775         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7776         internal_fn_mask_index and internal_gather_scatter_fn_p.
7777         (check_load_store_masking): Take the gather_scatter_info as an
7778         argument and handle gather loads.
7779         (vect_get_gather_scatter_ops): New function.
7780         (vectorizable_call): Check internal_load_fn_p.
7781         (vectorizable_load): Likewise.  Handle gather load internal
7782         functions.
7783         (vectorizable_store): Update call to check_load_store_masking.
7784         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
7785         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
7786         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
7787         (aarch64_gather_scale_operand_d): New predicates.
7788         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
7789         (mask_gather_load<mode>): New insns.
7790
7791 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7792             Alan Hayward  <alan.hayward@arm.com>
7793             David Sherwood  <david.sherwood@arm.com>
7794
7795         * optabs.def (fold_left_plus_optab): New optab.
7796         * doc/md.texi (fold_left_plus_@var{m}): Document.
7797         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
7798         * internal-fn.c (fold_left_direct): Define.
7799         (expand_fold_left_optab_fn): Likewise.
7800         (direct_fold_left_optab_supported_p): Likewise.
7801         * fold-const-call.c (fold_const_fold_left): New function.
7802         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
7803         * tree-parloops.c (valid_reduction_p): New function.
7804         (gather_scalar_reductions): Use it.
7805         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
7806         (vect_finish_replace_stmt): Declare.
7807         * tree-vect-loop.c (fold_left_reduction_fn): New function.
7808         (needs_fold_left_reduction_p): New function, split out from...
7809         (vect_is_simple_reduction): ...here.  Accept reductions that
7810         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
7811         (vect_force_simple_reduction): Also store the reduction type in
7812         the assignment's STMT_VINFO_REDUC_TYPE.
7813         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
7814         (merge_with_identity): New function.
7815         (vect_expand_fold_left): Likewise.
7816         (vectorize_fold_left_reduction): Likewise.
7817         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
7818         scalar phi in place for it.  Check for target support and reject
7819         cases that would reassociate the operation.  Defer the transform
7820         phase to vectorize_fold_left_reduction.
7821         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
7822         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
7823         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
7824
7825 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7826
7827         * tree-if-conv.c (predicate_mem_writes): Remove redundant
7828         call to ifc_temp_var.
7829
7830 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7831             Alan Hayward  <alan.hayward@arm.com>
7832             David Sherwood  <david.sherwood@arm.com>
7833
7834         * target.def (legitimize_address_displacement): Take the original
7835         offset as a poly_int.
7836         * targhooks.h (default_legitimize_address_displacement): Update
7837         accordingly.
7838         * targhooks.c (default_legitimize_address_displacement): Likewise.
7839         * doc/tm.texi: Regenerate.
7840         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
7841         as an argument, moving assert of ad->disp == ad->disp_term to...
7842         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
7843         Try calling targetm.legitimize_address_displacement before expanding
7844         the address rather than afterwards, and adjust for the new interface.
7845         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
7846         Match the new hook interface.  Handle SVE addresses.
7847         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
7848         new hook interface.
7849
7850 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7851
7852         * Makefile.in (OBJS): Add early-remat.o.
7853         * target.def (select_early_remat_modes): New hook.
7854         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
7855         * doc/tm.texi: Regenerate.
7856         * targhooks.h (default_select_early_remat_modes): Declare.
7857         * targhooks.c (default_select_early_remat_modes): New function.
7858         * timevar.def (TV_EARLY_REMAT): New timevar.
7859         * passes.def (pass_early_remat): New pass.
7860         * tree-pass.h (make_pass_early_remat): Declare.
7861         * early-remat.c: New file.
7862         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
7863         function.
7864         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
7865
7866 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7867             Alan Hayward  <alan.hayward@arm.com>
7868             David Sherwood  <david.sherwood@arm.com>
7869
7870         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
7871         vfm1 with a bound_epilog parameter.
7872         (vect_do_peeling): Update calls accordingly, and move the prologue
7873         call earlier in the function.  Treat the base bound_epilog as 0 for
7874         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
7875         this base when peeling for gaps.
7876         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
7877         with fully-masked loops.
7878         (vect_estimate_min_profitable_iters): Handle the single peeled
7879         iteration in that case.
7880
7881 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7882             Alan Hayward  <alan.hayward@arm.com>
7883             David Sherwood  <david.sherwood@arm.com>
7884
7885         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
7886         single-element interleaving even if the size is not a power of 2.
7887         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
7888         accesses for single-element interleaving if the group size is
7889         not a power of 2.
7890
7891 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7892             Alan Hayward  <alan.hayward@arm.com>
7893             David Sherwood  <david.sherwood@arm.com>
7894
7895         * doc/md.texi (fold_extract_last_@var{m}): Document.
7896         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
7897         * optabs.def (fold_extract_last_optab): New optab.
7898         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
7899         * internal-fn.c (fold_extract_direct): New macro.
7900         (expand_fold_extract_optab_fn): Likewise.
7901         (direct_fold_extract_optab_supported_p): Likewise.
7902         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
7903         * tree-vect-loop.c (vect_model_reduction_cost): Handle
7904         EXTRACT_LAST_REDUCTION.
7905         (get_initial_def_for_reduction): Do not create an initial vector
7906         for EXTRACT_LAST_REDUCTION reductions.
7907         (vectorizable_reduction): Leave the scalar phi in place for
7908         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
7909         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
7910         epilogue code for EXTRACT_LAST_REDUCTION and defer the
7911         transform phase to vectorizable_condition.
7912         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
7913         split out from...
7914         (vect_finish_stmt_generation): ...here.
7915         (vect_finish_replace_stmt): New function.
7916         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
7917         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
7918         pattern.
7919         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
7920
7921 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7922             Alan Hayward  <alan.hayward@arm.com>
7923             David Sherwood  <david.sherwood@arm.com>
7924
7925         * doc/md.texi (extract_last_@var{m}): Document.
7926         * optabs.def (extract_last_optab): New optab.
7927         * internal-fn.def (EXTRACT_LAST): New internal function.
7928         * internal-fn.c (cond_unary_direct): New macro.
7929         (expand_cond_unary_optab_fn): Likewise.
7930         (direct_cond_unary_optab_supported_p): Likewise.
7931         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
7932         loops using EXTRACT_LAST.
7933         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
7934         (extract_last_<mode>): ...this optab.
7935         (vec_extract<mode><Vel>): Update accordingly.
7936
7937 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7938             Alan Hayward  <alan.hayward@arm.com>
7939             David Sherwood  <david.sherwood@arm.com>
7940
7941         * target.def (empty_mask_is_expensive): New hook.
7942         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
7943         * doc/tm.texi: Regenerate.
7944         * targhooks.h (default_empty_mask_is_expensive): Declare.
7945         * targhooks.c (default_empty_mask_is_expensive): New function.
7946         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
7947         if the target says that empty masks are expensive.
7948         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
7949         New function.
7950         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
7951
7952 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7953             Alan Hayward  <alan.hayward@arm.com>
7954             David Sherwood  <david.sherwood@arm.com>
7955
7956         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
7957         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
7958         (vect_use_loop_mask_for_alignment_p): New function.
7959         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
7960         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
7961         niters_skip argument.  Make sure that the first niters_skip elements
7962         of the first iteration are inactive.
7963         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
7964         Update call to vect_set_loop_masks_directly.
7965         (get_misalign_in_elems): New function, split out from...
7966         (vect_gen_prolog_loop_niters): ...here.
7967         (vect_update_init_of_dr): Take a code argument that specifies whether
7968         the adjustment should be added or subtracted.
7969         (vect_update_init_of_drs): Likewise.
7970         (vect_prepare_for_masked_peels): New function.
7971         (vect_do_peeling): Skip prologue peeling if we're using a mask
7972         instead.  Update call to vect_update_inits_of_drs.
7973         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
7974         mask_skip_niters.
7975         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
7976         alignment.  Do not include the number of peeled iterations in
7977         the minimum threshold in that case.
7978         (vectorizable_induction): Adjust the start value down by
7979         LOOP_VINFO_MASK_SKIP_NITERS iterations.
7980         (vect_transform_loop): Call vect_prepare_for_masked_peels.
7981         Take the number of skipped iterations into account when calculating
7982         the loop bounds.
7983         * tree-vect-stmts.c (vect_gen_while_not): New function.
7984
7985 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7986             Alan Hayward  <alan.hayward@arm.com>
7987             David Sherwood  <david.sherwood@arm.com>
7988
7989         * doc/sourcebuild.texi (vect_fully_masked): Document.
7990         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
7991         default value to 0.
7992         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
7993         split out from...
7994         (vect_analyze_loop_2): ...here. Don't check the vectorization
7995         factor against the number of loop iterations if the loop is
7996         fully-masked.
7997
7998 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7999             Alan Hayward  <alan.hayward@arm.com>
8000             David Sherwood  <david.sherwood@arm.com>
8001
8002         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
8003         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
8004         (dump_groups): Update accordingly.
8005         (iv_use::mem_type): New member variable.
8006         (address_p): New function.
8007         (record_use): Add a mem_type argument and initialize the new
8008         mem_type field.
8009         (record_group_use): Add a mem_type argument.  Use address_p.
8010         Remove obsolete null checks of base_object.  Update call to record_use.
8011         (find_interesting_uses_op): Update call to record_group_use.
8012         (find_interesting_uses_cond): Likewise.
8013         (find_interesting_uses_address): Likewise.
8014         (get_mem_type_for_internal_fn): New function.
8015         (find_address_like_use): Likewise.
8016         (find_interesting_uses_stmt): Try find_address_like_use before
8017         calling find_interesting_uses_op.
8018         (addr_offset_valid_p): Use the iv mem_type field as the type
8019         of the addressed memory.
8020         (add_autoinc_candidates): Likewise.
8021         (get_address_cost): Likewise.
8022         (split_small_address_groups_p): Use address_p.
8023         (split_address_groups): Likewise.
8024         (add_iv_candidate_for_use): Likewise.
8025         (autoinc_possible_for_pair): Likewise.
8026         (rewrite_groups): Likewise.
8027         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
8028         (determine_group_iv_cost): Update after split of USE_ADDRESS.
8029         (get_alias_ptr_type_for_ptr_address): New function.
8030         (rewrite_use_address): Rewrite address uses in calls that were
8031         identified by find_address_like_use.
8032
8033 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8034             Alan Hayward  <alan.hayward@arm.com>
8035             David Sherwood  <david.sherwood@arm.com>
8036
8037         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
8038         TARGET_MEM_REFs.
8039         * gimple-expr.h (is_gimple_addressable: Likewise.
8040         * gimple-expr.c (is_gimple_address): Likewise.
8041         * internal-fn.c (expand_call_mem_ref): New function.
8042         (expand_mask_load_optab_fn): Use it.
8043         (expand_mask_store_optab_fn): Likewise.
8044
8045 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8046             Alan Hayward  <alan.hayward@arm.com>
8047             David Sherwood  <david.sherwood@arm.com>
8048
8049         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
8050         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
8051         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
8052         (cond_umax@var{mode}): Document.
8053         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
8054         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
8055         (cond_umin_optab, cond_umax_optab): New optabs.
8056         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
8057         (COND_IOR, COND_XOR): New internal functions.
8058         * internal-fn.h (get_conditional_internal_fn): Declare.
8059         * internal-fn.c (cond_binary_direct): New macro.
8060         (expand_cond_binary_optab_fn): Likewise.
8061         (direct_cond_binary_optab_supported_p): Likewise.
8062         (get_conditional_internal_fn): New function.
8063         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
8064         Cope with reduction statements that are vectorized as calls rather
8065         than assignments.
8066         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
8067         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
8068         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
8069         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
8070         (UNSPEC_COND_EOR): New unspecs.
8071         (optab): Add mappings for them.
8072         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
8073         (sve_int_op, sve_fp_op): New int attributes.
8074
8075 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8076             Alan Hayward  <alan.hayward@arm.com>
8077             David Sherwood  <david.sherwood@arm.com>
8078
8079         * optabs.def (while_ult_optab): New optab.
8080         * doc/md.texi (while_ult@var{m}@var{n}): Document.
8081         * internal-fn.def (WHILE_ULT): New internal function.
8082         * internal-fn.h (direct_internal_fn_supported_p): New override
8083         that takes two types as argument.
8084         * internal-fn.c (while_direct): New macro.
8085         (expand_while_optab_fn): New function.
8086         (convert_optab_supported_p): Likewise.
8087         (direct_while_optab_supported_p): New macro.
8088         * wide-int.h (wi::udiv_ceil): New function.
8089         * tree-vectorizer.h (rgroup_masks): New structure.
8090         (vec_loop_masks): New typedef.
8091         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
8092         and fully_masked_p.
8093         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
8094         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
8095         (vect_max_vf): New function.
8096         (slpeel_make_loop_iterate_ntimes): Delete.
8097         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
8098         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
8099         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
8100         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
8101         internal-fn.h, stor-layout.h and optabs-query.h.
8102         (vect_set_loop_mask): New function.
8103         (add_preheader_seq): Likewise.
8104         (add_header_seq): Likewise.
8105         (interleave_supported_p): Likewise.
8106         (vect_maybe_permute_loop_masks): Likewise.
8107         (vect_set_loop_masks_directly): Likewise.
8108         (vect_set_loop_condition_masked): Likewise.
8109         (vect_set_loop_condition_unmasked): New function, split out from
8110         slpeel_make_loop_iterate_ntimes.
8111         (slpeel_make_loop_iterate_ntimes): Rename to..
8112         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
8113         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
8114         (vect_do_peeling): Update call accordingly.
8115         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
8116         loops.
8117         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
8118         mask_compare_type, can_fully_mask_p and fully_masked_p.
8119         (release_vec_loop_masks): New function.
8120         (_loop_vec_info): Use it to free the loop masks.
8121         (can_produce_all_loop_masks_p): New function.
8122         (vect_get_max_nscalars_per_iter): Likewise.
8123         (vect_verify_full_masking): Likewise.
8124         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
8125         retries, and free the mask rgroups before retrying.  Check loop-wide
8126         reasons for disallowing fully-masked loops.  Make the final decision
8127         about whether use a fully-masked loop or not.
8128         (vect_estimate_min_profitable_iters): Do not assume that peeling
8129         for the number of iterations will be needed for fully-masked loops.
8130         (vectorizable_reduction): Disable fully-masked loops.
8131         (vectorizable_live_operation): Likewise.
8132         (vect_halve_mask_nunits): New function.
8133         (vect_double_mask_nunits): Likewise.
8134         (vect_record_loop_mask): Likewise.
8135         (vect_get_loop_mask): Likewise.
8136         (vect_transform_loop): Handle the case in which the final loop
8137         iteration might handle a partial vector.  Call vect_set_loop_condition
8138         instead of slpeel_make_loop_iterate_ntimes.
8139         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
8140         (check_load_store_masking): New function.
8141         (prepare_load_store_mask): Likewise.
8142         (vectorizable_store): Handle fully-masked loops.
8143         (vectorizable_load): Likewise.
8144         (supportable_widening_operation): Use vect_halve_mask_nunits for
8145         booleans.
8146         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
8147         (vect_gen_while): New function.
8148         * config/aarch64/aarch64.md (umax<mode>3): New expander.
8149         (aarch64_uqdec<mode>): New insn.
8150
8151 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8152             Alan Hayward  <alan.hayward@arm.com>
8153             David Sherwood  <david.sherwood@arm.com>
8154
8155         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
8156         (reduc_xor_scal_optab): New optabs.
8157         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
8158         (reduc_xor_scal_@var{m}): Document.
8159         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
8160         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
8161         internal functions.
8162         * fold-const-call.c (fold_const_call): Handle them.
8163         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
8164         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
8165         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
8166         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
8167         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
8168         (UNSPEC_XORV): New unspecs.
8169         (optab): Add entries for them.
8170         (BITWISEV): New int iterator.
8171         (bit_reduc_op): New int attributes.
8172
8173 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8174             Alan Hayward  <alan.hayward@arm.com>
8175             David Sherwood  <david.sherwood@arm.com>
8176
8177         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
8178         * internal-fn.def (VEC_SHL_INSERT): New internal function.
8179         * optabs.def (vec_shl_insert_optab): New optab.
8180         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
8181         (duplicate_and_interleave): Likewise.
8182         * tree-vect-loop.c: Include internal-fn.h.
8183         (neutral_op_for_slp_reduction): New function, split out from
8184         get_initial_defs_for_reduction.
8185         (get_initial_def_for_reduction): Handle option 2 for variable-length
8186         vectors by loading the neutral value into a vector and then shifting
8187         the initial value into element 0.
8188         (get_initial_defs_for_reduction): Replace the code argument with
8189         the neutral value calculated by neutral_op_for_slp_reduction.
8190         Use gimple_build_vector for constant-length vectors.
8191         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
8192         but the first group_size elements have a neutral value.
8193         Use duplicate_and_interleave otherwise.
8194         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
8195         Update call to get_initial_defs_for_reduction.  Handle SLP
8196         reductions for variable-length vectors by creating one vector
8197         result for each scalar result, with the elements associated
8198         with other scalar results stubbed out with the neutral value.
8199         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
8200         Require IFN_VEC_SHL_INSERT for double reductions on
8201         variable-length vectors, or SLP reductions that have
8202         a neutral value.  Require can_duplicate_and_interleave_p
8203         support for variable-length unchained SLP reductions if there
8204         is no neutral value, such as for MIN/MAX reductions.  Also require
8205         the number of vector elements to be a multiple of the number of
8206         SLP statements when doing variable-length unchained SLP reductions.
8207         Update call to vect_create_epilog_for_reduction.
8208         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
8209         and remove initial values.
8210         (duplicate_and_interleave): Make public.
8211         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
8212         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
8213
8214 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8215             Alan Hayward  <alan.hayward@arm.com>
8216             David Sherwood  <david.sherwood@arm.com>
8217
8218         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
8219         (can_duplicate_and_interleave_p): New function.
8220         (vect_get_and_check_slp_defs): Take the vector of statements
8221         rather than just the current one.  Remove excess parentheses.
8222         Restriction rejectinon of vect_constant_def and vect_external_def
8223         for variable-length vectors to boolean types, or types for which
8224         can_duplicate_and_interleave_p is false.
8225         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
8226         (duplicate_and_interleave): New function.
8227         (vect_get_constant_vectors): Use gimple_build_vector for
8228         constant-length vectors and suitable variable-length constant
8229         vectors.  Use duplicate_and_interleave for other variable-length
8230         vectors.  Don't defer the update when inserting new statements.
8231
8232 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8233             Alan Hayward  <alan.hayward@arm.com>
8234             David Sherwood  <david.sherwood@arm.com>
8235
8236         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
8237         min_profitable_iters doesn't go negative.
8238
8239 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8240             Alan Hayward  <alan.hayward@arm.com>
8241             David Sherwood  <david.sherwood@arm.com>
8242
8243         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
8244         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
8245         * optabs.def (vec_mask_load_lanes_optab): New optab.
8246         (vec_mask_store_lanes_optab): Likewise.
8247         * internal-fn.def (MASK_LOAD_LANES): New internal function.
8248         (MASK_STORE_LANES): Likewise.
8249         * internal-fn.c (mask_load_lanes_direct): New macro.
8250         (mask_store_lanes_direct): Likewise.
8251         (expand_mask_load_optab_fn): Handle masked operations.
8252         (expand_mask_load_lanes_optab_fn): New macro.
8253         (expand_mask_store_optab_fn): Handle masked operations.
8254         (expand_mask_store_lanes_optab_fn): New macro.
8255         (direct_mask_load_lanes_optab_supported_p): Likewise.
8256         (direct_mask_store_lanes_optab_supported_p): Likewise.
8257         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
8258         parameter.
8259         (vect_load_lanes_supported): Likewise.
8260         * tree-vect-data-refs.c (strip_conversion): New function.
8261         (can_group_stmts_p): Likewise.
8262         (vect_analyze_data_ref_accesses): Use it instead of checking
8263         for a pair of assignments.
8264         (vect_store_lanes_supported): Take a masked_p parameter.
8265         (vect_load_lanes_supported): Likewise.
8266         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
8267         vect_store_lanes_supported and vect_load_lanes_supported.
8268         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
8269         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
8270         parameter.  Don't allow gaps for masked accesses.
8271         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
8272         and vect_load_lanes_supported.
8273         (get_load_store_type): Take a masked_p parameter and update
8274         call to get_group_load_store_type.
8275         (vectorizable_store): Update call to get_load_store_type.
8276         Handle IFN_MASK_STORE_LANES.
8277         (vectorizable_load): Update call to get_load_store_type.
8278         Handle IFN_MASK_LOAD_LANES.
8279
8280 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8281             Alan Hayward  <alan.hayward@arm.com>
8282             David Sherwood  <david.sherwood@arm.com>
8283
8284         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
8285         modes for SVE.
8286         * config/aarch64/aarch64-protos.h
8287         (aarch64_sve_struct_memory_operand_p): Declare.
8288         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
8289         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
8290         (VPRED, vpred): Handle SVE structure modes.
8291         * config/aarch64/constraints.md (Utx): New constraint.
8292         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
8293         (aarch64_sve_struct_nonimmediate_operand): New predicates.
8294         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
8295         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
8296         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
8297         structure modes.  Split into pieces after RA.
8298         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
8299         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
8300         New patterns.
8301         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
8302         SVE structure modes.
8303         (aarch64_classify_address): Likewise.
8304         (sizetochar): Move earlier in file.
8305         (aarch64_print_operand): Handle SVE register lists.
8306         (aarch64_array_mode): New function.
8307         (aarch64_sve_struct_memory_operand_p): Likewise.
8308         (TARGET_ARRAY_MODE): Redefine.
8309
8310 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8311             Alan Hayward  <alan.hayward@arm.com>
8312             David Sherwood  <david.sherwood@arm.com>
8313
8314         * target.def (array_mode): New target hook.
8315         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
8316         * doc/tm.texi: Regenerate.
8317         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
8318         * hooks.c (hook_optmode_mode_uhwi_none): New function.
8319         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
8320         targetm.array_mode.
8321         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
8322         type sizes.
8323
8324 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8325             Alan Hayward  <alan.hayward@arm.com>
8326             David Sherwood  <david.sherwood@arm.com>
8327
8328         * fold-const.c (fold_binary_loc): Check the argument types
8329         rather than the result type when testing for a vector operation.
8330
8331 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8332
8333         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
8334         * doc/tm.texi: Regenerate.
8335
8336 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8337             Alan Hayward  <alan.hayward@arm.com>
8338             David Sherwood  <david.sherwood@arm.com>
8339
8340         * doc/invoke.texi (-msve-vector-bits=): Document new option.
8341         (sve): Document new AArch64 extension.
8342         * doc/md.texi (w): Extend the description of the AArch64
8343         constraint to include SVE vectors.
8344         (Upl, Upa): Document new AArch64 predicate constraints.
8345         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
8346         enum.
8347         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
8348         (msve-vector-bits=): New option.
8349         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
8350         SVE when these are disabled.
8351         (sve): New extension.
8352         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
8353         modes.  Adjust their number of units based on aarch64_sve_vg.
8354         (MAX_BITSIZE_MODE_ANY_MODE): Define.
8355         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
8356         aarch64_addr_query_type.
8357         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
8358         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
8359         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
8360         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
8361         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
8362         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
8363         (aarch64_simd_imm_zero_p): Delete.
8364         (aarch64_check_zero_based_sve_index_immediate): Declare.
8365         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8366         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8367         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8368         (aarch64_sve_float_mul_immediate_p): Likewise.
8369         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8370         rather than an rtx.
8371         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
8372         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
8373         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
8374         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
8375         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
8376         (aarch64_regmode_natural_size): Likewise.
8377         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
8378         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
8379         left one place.
8380         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
8381         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
8382         for VG and the SVE predicate registers.
8383         (V_ALIASES): Add a "z"-prefixed alias.
8384         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
8385         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
8386         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
8387         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
8388         (REG_CLASS_NAMES): Add entries for them.
8389         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
8390         and the predicate registers.
8391         (aarch64_sve_vg): Declare.
8392         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
8393         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
8394         (REGMODE_NATURAL_SIZE): Define.
8395         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
8396         SVE macros.
8397         * config/aarch64/aarch64.c: Include cfgrtl.h.
8398         (simd_immediate_info): Add a constructor for series vectors,
8399         and an associated step field.
8400         (aarch64_sve_vg): New variable.
8401         (aarch64_dbx_register_number): Handle VG and the predicate registers.
8402         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
8403         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
8404         (VEC_ANY_DATA, VEC_STRUCT): New constants.
8405         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
8406         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
8407         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
8408         (aarch64_get_mask_mode): New functions.
8409         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
8410         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
8411         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
8412         predicate modes and predicate registers.  Explicitly restrict
8413         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
8414         to store a vector mode if it is recognized by
8415         aarch64_classify_vector_mode.
8416         (aarch64_regmode_natural_size): New function.
8417         (aarch64_hard_regno_caller_save_mode): Return the original mode
8418         for predicates.
8419         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
8420         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
8421         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
8422         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
8423         functions.
8424         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
8425         does not overlap dest if the function is frame-related.  Handle
8426         SVE constants.
8427         (aarch64_split_add_offset): New function.
8428         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
8429         them aarch64_add_offset.
8430         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
8431         and update call to aarch64_sub_sp.
8432         (aarch64_add_cfa_expression): New function.
8433         (aarch64_expand_prologue): Pass extra temporary registers to the
8434         functions above.  Handle the case in which we need to emit new
8435         DW_CFA_expressions for registers that were originally saved
8436         relative to the stack pointer, but now have to be expressed
8437         relative to the frame pointer.
8438         (aarch64_output_mi_thunk): Pass extra temporary registers to the
8439         functions above.
8440         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
8441         IP0 and IP1 values for SVE frames.
8442         (aarch64_expand_vec_series): New function.
8443         (aarch64_expand_sve_widened_duplicate): Likewise.
8444         (aarch64_expand_sve_const_vector): Likewise.
8445         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
8446         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
8447         into the register, rather than emitting a SET directly.
8448         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
8449         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
8450         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
8451         (offset_9bit_signed_scaled_p): New functions.
8452         (aarch64_replicate_bitmask_imm): New function.
8453         (aarch64_bitmask_imm): Use it.
8454         (aarch64_cannot_force_const_mem): Reject expressions involving
8455         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
8456         (aarch64_classify_index): Handle SVE indices, by requiring
8457         a plain register index with a scale that matches the element size.
8458         (aarch64_classify_address): Handle SVE addresses.  Assert that
8459         the mode of the address is VOIDmode or an integer mode.
8460         Update call to aarch64_classify_symbol.
8461         (aarch64_classify_symbolic_expression): Update call to
8462         aarch64_classify_symbol.
8463         (aarch64_const_vec_all_in_range_p): New function.
8464         (aarch64_print_vector_float_operand): Likewise.
8465         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
8466         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
8467         and the FP immediates 1.0 and 0.5.
8468         (aarch64_print_address_internal): Handle SVE addresses.
8469         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
8470         (aarch64_regno_regclass): Handle predicate registers.
8471         (aarch64_secondary_reload): Handle big-endian reloads of SVE
8472         data modes.
8473         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
8474         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
8475         (aarch64_convert_sve_vector_bits): New function.
8476         (aarch64_override_options): Use it to handle -msve-vector-bits=.
8477         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8478         rather than an rtx.
8479         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
8480         Handle SVE vector and predicate modes.  Accept VL-based constants
8481         that need only one temporary register, and VL offsets that require
8482         no temporary registers.
8483         (aarch64_conditional_register_usage): Mark the predicate registers
8484         as fixed if SVE isn't available.
8485         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
8486         Return true for SVE vector and predicate modes.
8487         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
8488         rather than an unsigned int.  Handle SVE modes.
8489         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
8490         SVE modes.
8491         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
8492         if SVE is enabled.
8493         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8494         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8495         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8496         (aarch64_sve_float_mul_immediate_p): New functions.
8497         (aarch64_sve_valid_immediate): New function.
8498         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
8499         Explicitly reject structure modes.  Check for INDEX constants.
8500         Handle PTRUE and PFALSE constants.
8501         (aarch64_check_zero_based_sve_index_immediate): New function.
8502         (aarch64_simd_imm_zero_p): Delete.
8503         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
8504         vector modes.  Accept constants in the range of CNT[BHWD].
8505         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
8506         ask for an Advanced SIMD mode.
8507         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
8508         (aarch64_simd_vector_alignment): Handle SVE predicates.
8509         (aarch64_vectorize_preferred_vector_alignment): New function.
8510         (aarch64_simd_vector_alignment_reachable): Use it instead of
8511         the vector size.
8512         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
8513         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
8514         functions.
8515         (MAX_VECT_LEN): Delete.
8516         (expand_vec_perm_d): Add a vec_flags field.
8517         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
8518         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
8519         (aarch64_evpc_ext): Don't apply a big-endian lane correction
8520         for SVE modes.
8521         (aarch64_evpc_rev): Rename to...
8522         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
8523         (aarch64_evpc_rev_global): New function.
8524         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
8525         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
8526         MAX_VECT_LEN.
8527         (aarch64_evpc_sve_tbl): New function.
8528         (aarch64_expand_vec_perm_const_1): Update after rename of
8529         aarch64_evpc_rev.  Handle SVE permutes too, trying
8530         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
8531         than aarch64_evpc_tbl.
8532         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
8533         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
8534         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
8535         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
8536         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
8537         (aarch64_expand_sve_vcond): New functions.
8538         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
8539         of aarch64_vector_mode_p.
8540         (aarch64_dwarf_poly_indeterminate_value): New function.
8541         (aarch64_compute_pressure_classes): Likewise.
8542         (aarch64_can_change_mode_class): Likewise.
8543         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
8544         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
8545         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
8546         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
8547         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
8548         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
8549         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
8550         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
8551         constraints.
8552         (Dn, Dl, Dr): Accept const as well as const_vector.
8553         (Dz): Likewise.  Compare against CONST0_RTX.
8554         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
8555         of "vector" where appropriate.
8556         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
8557         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
8558         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
8559         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
8560         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
8561         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
8562         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
8563         (v_int_equiv): Extend to SVE modes.
8564         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
8565         mode attributes.
8566         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
8567         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
8568         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
8569         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
8570         (SVE_COND_FP_CMP): New int iterators.
8571         (perm_hilo): Handle the new unpack unspecs.
8572         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
8573         attributes.
8574         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
8575         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
8576         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
8577         (aarch64_equality_operator, aarch64_constant_vector_operand)
8578         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
8579         (aarch64_sve_nonimmediate_operand): Likewise.
8580         (aarch64_sve_general_operand): Likewise.
8581         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
8582         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
8583         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
8584         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
8585         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
8586         (aarch64_sve_float_arith_immediate): Likewise.
8587         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
8588         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
8589         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
8590         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
8591         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
8592         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
8593         (aarch64_sve_float_arith_operand): Likewise.
8594         (aarch64_sve_float_arith_with_sub_operand): Likewise.
8595         (aarch64_sve_float_mul_operand): Likewise.
8596         (aarch64_sve_vec_perm_operand): Likewise.
8597         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
8598         (aarch64_mov_operand): Accept const_poly_int and const_vector.
8599         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
8600         as well as const_vector.
8601         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
8602         in file.  Use CONST0_RTX and CONSTM1_RTX.
8603         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
8604         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
8605         Use aarch64_simd_imm_zero.
8606         * config/aarch64/aarch64-sve.md: New file.
8607         * config/aarch64/aarch64.md: Include it.
8608         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
8609         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
8610         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
8611         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
8612         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
8613         (sve): New attribute.
8614         (enabled): Disable instructions with the sve attribute unless
8615         TARGET_SVE.
8616         (movqi, movhi): Pass CONST_POLY_INT operaneds through
8617         aarch64_expand_mov_immediate.
8618         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
8619         CNT[BHSD] immediates.
8620         (movti): Split CONST_POLY_INT moves into two halves.
8621         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
8622         Split additions that need a temporary here if the destination
8623         is the stack pointer.
8624         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
8625         (*add<mode>3_poly_1): New instruction.
8626         (set_clobber_cc): New expander.
8627
8628 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8629
8630         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
8631         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
8632         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
8633         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
8634         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
8635         Change innermode from fixed_mode_size to machine_mode.
8636         (simplify_subreg): Update call accordingly.  Handle a constant-sized
8637         subreg of a variable-length CONST_VECTOR.
8638
8639 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8640             Alan Hayward  <alan.hayward@arm.com>
8641             David Sherwood  <david.sherwood@arm.com>
8642
8643         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
8644         (add_offset_to_base): New function, split out from...
8645         (create_mem_ref): ...here.  When handling a scale other than 1,
8646         check first whether the address is valid without the offset.
8647         Add it into the base if so, leaving the index and scale as-is.
8648
8649 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8650
8651         PR c++/83778
8652         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
8653         fold_for_warn before checking if arg2 is INTEGER_CST.
8654
8655 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
8656
8657         * config/rs6000/predicates.md (load_multiple_operation): Delete.
8658         (store_multiple_operation): Delete.
8659         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
8660         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
8661         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
8662         guarded by TARGET_STRING.
8663         (rs6000_output_load_multiple): Delete.
8664         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
8665         OPTION_MASK_STRING / TARGET_STRING handling.
8666         (print_operand) <'N', 'O'>: Add comment that these are unused now.
8667         (const rs6000_opt_masks) <"string">: Change mask to 0.
8668         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
8669         (MASK_STRING): Delete.
8670         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
8671         parts.  Simplify.
8672         (load_multiple): Delete.
8673         (*ldmsi8): Delete.
8674         (*ldmsi7): Delete.
8675         (*ldmsi6): Delete.
8676         (*ldmsi5): Delete.
8677         (*ldmsi4): Delete.
8678         (*ldmsi3): Delete.
8679         (store_multiple): Delete.
8680         (*stmsi8): Delete.
8681         (*stmsi7): Delete.
8682         (*stmsi6): Delete.
8683         (*stmsi5): Delete.
8684         (*stmsi4): Delete.
8685         (*stmsi3): Delete.
8686         (movmemsi_8reg): Delete.
8687         (corresponding unnamed define_insn): Delete.
8688         (movmemsi_6reg): Delete.
8689         (corresponding unnamed define_insn): Delete.
8690         (movmemsi_4reg): Delete.
8691         (corresponding unnamed define_insn): Delete.
8692         (movmemsi_2reg): Delete.
8693         (corresponding unnamed define_insn): Delete.
8694         (movmemsi_1reg): Delete.
8695         (corresponding unnamed define_insn): Delete.
8696         * config/rs6000/rs6000.opt (mno-string): New.
8697         (mstring): Replace by deprecation warning stub.
8698         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
8699
8700 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8701
8702         * regrename.c (regrename_do_replace): If replacing the same
8703         reg multiple times, try to reuse last created gen_raw_REG.
8704
8705         PR debug/81155
8706         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
8707         main to workaround a bug in GDB.
8708
8709 2018-01-12  Tom de Vries  <tom@codesourcery.com>
8710
8711         PR target/83737
8712         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
8713
8714 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
8715
8716         PR rtl-optimization/80481
8717         * ira-color.c (get_cap_member): New function.
8718         (allocnos_conflict_by_live_ranges_p): Use it.
8719         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
8720         (setup_slot_coalesced_allocno_live_ranges): Ditto.
8721
8722 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
8723
8724         PR target/83628
8725         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
8726         (*saddl_se_1): Ditto.
8727         (*ssubsi_1): Ditto.
8728         (*ssubl_se_1): Ditto.
8729
8730 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8731
8732         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
8733         rather than wi::to_widest for DR_INITs.
8734         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
8735         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
8736         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
8737         INTEGER_CSTs.
8738         (vect_analyze_group_access_1): Note that here.
8739
8740 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8741
8742         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
8743         polynomial type sizes.
8744
8745 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8746
8747         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
8748         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
8749         (gimple_add_tmp_var): Likewise.
8750
8751 2018-01-12  Martin Liska  <mliska@suse.cz>
8752
8753         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
8754         (gimple_alloc_sizes): Likewise.
8755         (dump_gimple_statistics): Use PRIu64 in printf format.
8756         * gimple.h: Change uint64_t to int.
8757
8758 2018-01-12  Martin Liska  <mliska@suse.cz>
8759
8760         * tree-core.h: Use uint64_t instead of int.
8761         * tree.c (tree_node_counts): Likewise.
8762         (tree_node_sizes): Likewise.
8763         (dump_tree_statistics): Use PRIu64 in printf format.
8764
8765 2018-01-12  Martin Liska  <mliska@suse.cz>
8766
8767         * Makefile.in: As qsort_chk is implemented in vec.c, add
8768         vec.o to linkage of gencfn-macros.
8769         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
8770         passing the info to record_node_allocation_statistics.
8771         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
8772         and pass the info.
8773         * ggc-common.c (struct ggc_usage): Add operator== and use
8774         it in operator< and compare function.
8775         * mem-stats.h (struct mem_usage): Likewise.
8776         * vec.c (struct vec_usage): Remove operator< and compare
8777         function. Can be simply inherited.
8778
8779 2018-01-12  Martin Jambor  <mjambor@suse.cz>
8780
8781         PR target/81616
8782         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
8783         * tree-ssa-math-opts.c: Include domwalk.h.
8784         (convert_mult_to_fma_1): New function.
8785         (fma_transformation_info): New type.
8786         (fma_deferring_state): Likewise.
8787         (cancel_fma_deferring): New function.
8788         (result_of_phi): Likewise.
8789         (last_fma_candidate_feeds_initial_phi): Likewise.
8790         (convert_mult_to_fma): Added deferring logic, split actual
8791         transformation to convert_mult_to_fma_1.
8792         (math_opts_dom_walker): New type.
8793         (math_opts_dom_walker::after_dom_children): New method, body moved
8794         here from pass_optimize_widening_mul::execute, added deferring logic
8795         bits.
8796         (pass_optimize_widening_mul::execute): Moved most of code to
8797         math_opts_dom_walker::after_dom_children.
8798         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
8799         * config/i386/i386.c (ix86_option_override_internal): Added
8800         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
8801
8802 2018-01-12  Richard Biener  <rguenther@suse.de>
8803
8804         PR debug/83157
8805         * dwarf2out.c (gen_variable_die): Do not reset old_die for
8806         inline instance vars.
8807
8808 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
8809
8810         PR target/81819
8811         * config/rx/rx.c (rx_is_restricted_memory_address):
8812         Handle SUBREG case.
8813
8814 2018-01-12  Richard Biener  <rguenther@suse.de>
8815
8816         PR tree-optimization/80846
8817         * target.def (split_reduction): New target hook.
8818         * targhooks.c (default_split_reduction): New function.
8819         * targhooks.h (default_split_reduction): Declare.
8820         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
8821         target requests first reduce vectors by combining low and high
8822         parts.
8823         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
8824         (get_vectype_for_scalar_type_and_size): Export.
8825         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
8826         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
8827         * doc/tm.texi: Regenerate.
8828         * config/i386/i386.c (ix86_split_reduction): Implement
8829         TARGET_VECTORIZE_SPLIT_REDUCTION.
8830
8831 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8832
8833         PR target/83368
8834         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
8835         in PIC mode except for TARGET_VXWORKS_RTP.
8836         * config/sparc/sparc.c: Include cfgrtl.h.
8837         (TARGET_INIT_PIC_REG): Define.
8838         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
8839         (sparc_pic_register_p): New predicate.
8840         (sparc_legitimate_address_p): Use it.
8841         (sparc_legitimize_pic_address): Likewise.
8842         (sparc_delegitimize_address): Likewise.
8843         (sparc_mode_dependent_address_p): Likewise.
8844         (gen_load_pcrel_sym): Remove 4th parameter.
8845         (load_got_register): Adjust call to above.  Remove obsolete stuff.
8846         (sparc_expand_prologue): Do not call load_got_register here.
8847         (sparc_flat_expand_prologue): Likewise.
8848         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
8849         (sparc_use_pseudo_pic_reg): New function.
8850         (sparc_init_pic_reg): Likewise.
8851         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
8852         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
8853
8854 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
8855
8856         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
8857         Add item for branch_cost.
8858
8859 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8860
8861         PR rtl-optimization/83565
8862         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
8863         not extend the result to a larger mode for rotate operations.
8864         (num_sign_bit_copies1): Likewise.
8865
8866 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8867
8868         PR target/40411
8869         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
8870         -symbolic.
8871         Use values-Xc.o for -pedantic.
8872         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
8873
8874 2018-01-12  Martin Liska  <mliska@suse.cz>
8875
8876         PR ipa/83054
8877         * ipa-devirt.c (final_warning_record::grow_type_warnings):
8878         New function.
8879         (possible_polymorphic_call_targets): Use it.
8880         (ipa_devirt): Likewise.
8881
8882 2018-01-12  Martin Liska  <mliska@suse.cz>
8883
8884         * profile-count.h (enum profile_quality): Use 0 as invalid
8885         enum value of profile_quality.
8886
8887 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
8888
8889         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
8890         -mext-string options.
8891
8892 2018-01-12  Richard Biener  <rguenther@suse.de>
8893
8894         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
8895         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
8896         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
8897         Likewise.
8898         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
8899
8900 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
8901
8902         * configure.ac (--with-long-double-format): Add support for the
8903         configuration option to change the default long double format on
8904         PowerPC systems.
8905         * config.gcc (powerpc*-linux*-*): Likewise.
8906         * configure: Regenerate.
8907         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
8908         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
8909         used without modification.
8910
8911 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8912
8913         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
8914         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
8915         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
8916         MISC_BUILTIN_SPEC_BARRIER.
8917         (rs6000_init_builtins): Likewise.
8918         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
8919         enum value.
8920         (speculation_barrier): New define_insn.
8921         * doc/extend.texi: Document __builtin_speculation_barrier.
8922
8923 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8924
8925         PR target/83203
8926         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
8927         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
8928         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
8929         iterators.
8930         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
8931         integral modes instead of "ss" and "sd".
8932         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
8933         vectors with 32-bit and 64-bit elements.
8934         (vecdupssescalarmodesuffix): New mode attribute.
8935         (vec_dup<mode>): Use it.
8936
8937 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
8938
8939         PR target/83330
8940         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
8941         frame if argument is passed on stack.
8942
8943 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8944
8945         PR target/82682
8946         * ree.c (combine_reaching_defs): Optimize also
8947         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
8948         reg2=any_extend(exp); reg1=reg2;, formatting fix.
8949
8950 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
8951
8952         PR middle-end/83189
8953         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
8954
8955 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
8956
8957         PR middle-end/83718
8958         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
8959         after they are computed.
8960
8961 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
8962
8963         PR tree-optimization/83695
8964         * gimple-loop-linterchange.cc
8965         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
8966         reset cached scev information after interchange.
8967         (pass_linterchange::execute): Remove call to scev_reset_htab.
8968
8969 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8970
8971         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
8972         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
8973         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
8974         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
8975         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
8976         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
8977         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
8978         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
8979         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
8980         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
8981         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
8982         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
8983         (V_lane_reg): Likewise.
8984         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
8985         New define_expand.
8986         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
8987         (vfmal_lane_low<mode>_intrinsic,
8988         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
8989         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
8990         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
8991         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
8992         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
8993         vfmsl_lane_high<mode>_intrinsic): New define_insns.
8994
8995 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8996
8997         * config/arm/arm-cpus.in (fp16fml): New feature.
8998         (ALL_SIMD): Add fp16fml.
8999         (armv8.2-a): Add fp16fml as an option.
9000         (armv8.3-a): Likewise.
9001         (armv8.4-a): Add fp16fml as part of fp16.
9002         * config/arm/arm.h (TARGET_FP16FML): Define.
9003         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
9004         when appropriate.
9005         * config/arm/arm-modes.def (V2HF): Define.
9006         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
9007         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
9008         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
9009         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
9010         vfmsl_low, vfmsl_high): New set of builtins.
9011         * config/arm/iterators.md (PLUSMINUS): New code iterator.
9012         (vfml_op): New code attribute.
9013         (VFMLHALVES): New int iterator.
9014         (VFML, VFMLSEL): New mode attributes.
9015         (V_reg): Define mapping for V2HF.
9016         (V_hi, V_lo): New mode attributes.
9017         (VF_constraint): Likewise.
9018         (vfml_half, vfml_half_selector): New int attributes.
9019         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
9020         define_expand.
9021         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
9022         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
9023         New define_insn.
9024         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
9025         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
9026         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
9027         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
9028         documentation.
9029         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
9030         Document new effective target and option set.
9031
9032 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9033
9034         * config/arm/arm-cpus.in (armv8_4): New feature.
9035         (ARMv8_4a): New fgroup.
9036         (armv8.4-a): New arch.
9037         * config/arm/arm-tables.opt: Regenerate.
9038         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
9039         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
9040         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
9041         Add matching rules for -march=armv8.4-a and extensions.
9042         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
9043
9044 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
9045
9046         PR target/81821
9047         * config/rx/rx.md (BW): New mode attribute.
9048         (sync_lock_test_and_setsi): Add mode suffix to insn output.
9049
9050 2018-01-11  Richard Biener  <rguenther@suse.de>
9051
9052         PR tree-optimization/83435
9053         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
9054         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
9055         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
9056
9057 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9058             Alan Hayward  <alan.hayward@arm.com>
9059             David Sherwood  <david.sherwood@arm.com>
9060
9061         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
9062         field.
9063         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
9064         (aarch64_print_address_internal): Use it to check for a zero offset.
9065
9066 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9067             Alan Hayward  <alan.hayward@arm.com>
9068             David Sherwood  <david.sherwood@arm.com>
9069
9070         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
9071         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
9072         Return a poly_int64 rather than a HOST_WIDE_INT.
9073         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
9074         rather than a HOST_WIDE_INT.
9075         * config/aarch64/aarch64.h (aarch64_frame): Protect with
9076         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
9077         hard_fp_offset, frame_size, initial_adjust, callee_offset and
9078         final_offset from HOST_WIDE_INT to poly_int64.
9079         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
9080         to_constant when getting the number of units in an Advanced SIMD
9081         mode.
9082         (aarch64_builtin_vectorized_function): Check for a constant number
9083         of units.
9084         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
9085         GET_MODE_SIZE.
9086         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
9087         attribute instead of GET_MODE_NUNITS.
9088         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
9089         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
9090         GET_MODE_SIZE for fixed-size registers.
9091         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
9092         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
9093         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
9094         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
9095         (aarch64_print_operand, aarch64_print_address_internal)
9096         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
9097         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
9098         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
9099         Handle polynomial GET_MODE_SIZE.
9100         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
9101         wider than SImode without modification.
9102         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
9103         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
9104         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
9105         passing and returning SVE modes.
9106         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
9107         rather than GEN_INT.
9108         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
9109         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
9110         (aarch64_allocate_and_probe_stack_space): Likewise.
9111         (aarch64_layout_frame): Cope with polynomial offsets.
9112         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
9113         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
9114         polynomial offsets.
9115         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
9116         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
9117         poly_int64 rather than a HOST_WIDE_INT.
9118         (aarch64_get_separate_components, aarch64_process_components)
9119         (aarch64_expand_prologue, aarch64_expand_epilogue)
9120         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
9121         (aarch64_anchor_offset): New function, split out from...
9122         (aarch64_legitimize_address): ...here.
9123         (aarch64_builtin_vectorization_cost): Handle polynomial
9124         TYPE_VECTOR_SUBPARTS.
9125         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
9126         GET_MODE_NUNITS.
9127         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
9128         number of elements from the PARALLEL rather than the mode.
9129         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
9130         rather than GET_MODE_BITSIZE.
9131         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
9132         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
9133         (aarch64_expand_vec_perm_const_1): Handle polynomial
9134         d->perm.length () and d->perm elements.
9135         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
9136         Apply to_constant to d->perm elements.
9137         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
9138         polynomial CONST_VECTOR_NUNITS.
9139         (aarch64_move_pointer): Take amount as a poly_int64 rather
9140         than an int.
9141         (aarch64_progress_pointer): Avoid temporary variable.
9142         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
9143         the mode attribute instead of GET_MODE.
9144
9145 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9146             Alan Hayward  <alan.hayward@arm.com>
9147             David Sherwood  <david.sherwood@arm.com>
9148
9149         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
9150         x exists before using it.
9151         (aarch64_add_constant_internal): Rename to...
9152         (aarch64_add_offset_1): ...this.  Replace regnum with separate
9153         src and dest rtxes.  Handle the case in which they're different,
9154         including when the offset is zero.  Replace scratchreg with an rtx.
9155         Use 2 additions if there is no spare register into which we can
9156         move a 16-bit constant.
9157         (aarch64_add_constant): Delete.
9158         (aarch64_add_offset): Replace reg with separate src and dest
9159         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
9160         Use aarch64_add_offset_1.
9161         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
9162         an rtx rather than an int.  Take the delta as a poly_int64
9163         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
9164         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
9165         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
9166         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
9167         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
9168         and aarch64_add_sp.
9169         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
9170         aarch64_add_constant.
9171
9172 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9173
9174         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
9175         Use scalar_float_mode.
9176
9177 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9178
9179         * config/aarch64/aarch64-simd.md
9180         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
9181         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
9182         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
9183         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
9184         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
9185         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
9186         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
9187         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
9188         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
9189         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
9190
9191 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9192
9193         PR target/83514
9194         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
9195         targ_options->x_arm_arch_string is non NULL.
9196
9197 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
9198
9199         * config/aarch64/aarch64.h
9200         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
9201
9202 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
9203
9204         PR target/82096
9205         * expmed.c (emit_store_flag_force): Swap if const op0
9206         and change VOIDmode to mode of op0.
9207
9208 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9209
9210         PR rtl-optimization/83761
9211         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
9212         than bytes to mode_for_size.
9213
9214 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9215
9216         PR middle-end/83189
9217         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
9218         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
9219         profile.
9220
9221 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9222
9223         PR middle-end/83575
9224         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
9225         when in layout mode.
9226         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
9227         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
9228         partition fixup.
9229
9230 2018-01-10  Michael Collison  <michael.collison@arm.com>
9231
9232         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
9233         * config/aarch64/aarch64-option-extension.def: Add
9234         AARCH64_OPT_EXTENSION of 'fp16fml'.
9235         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9236         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
9237         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
9238         * config/aarch64/constraints.md (Ui7): New constraint.
9239         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
9240         (VFMLA_SEL_W): Ditto.
9241         (f16quad): Ditto.
9242         (f16mac1): Ditto.
9243         (VFMLA16_LOW): New int iterator.
9244         (VFMLA16_HIGH): Ditto.
9245         (UNSPEC_FMLAL): New unspec.
9246         (UNSPEC_FMLSL): Ditto.
9247         (UNSPEC_FMLAL2): Ditto.
9248         (UNSPEC_FMLSL2): Ditto.
9249         (f16mac): New code attribute.
9250         * config/aarch64/aarch64-simd-builtins.def
9251         (aarch64_fmlal_lowv2sf): Ditto.
9252         (aarch64_fmlsl_lowv2sf): Ditto.
9253         (aarch64_fmlalq_lowv4sf): Ditto.
9254         (aarch64_fmlslq_lowv4sf): Ditto.
9255         (aarch64_fmlal_highv2sf): Ditto.
9256         (aarch64_fmlsl_highv2sf): Ditto.
9257         (aarch64_fmlalq_highv4sf): Ditto.
9258         (aarch64_fmlslq_highv4sf): Ditto.
9259         (aarch64_fmlal_lane_lowv2sf): Ditto.
9260         (aarch64_fmlsl_lane_lowv2sf): Ditto.
9261         (aarch64_fmlal_laneq_lowv2sf): Ditto.
9262         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
9263         (aarch64_fmlalq_lane_lowv4sf): Ditto.
9264         (aarch64_fmlsl_lane_lowv4sf): Ditto.
9265         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
9266         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
9267         (aarch64_fmlal_lane_highv2sf): Ditto.
9268         (aarch64_fmlsl_lane_highv2sf): Ditto.
9269         (aarch64_fmlal_laneq_highv2sf): Ditto.
9270         (aarch64_fmlsl_laneq_highv2sf): Ditto.
9271         (aarch64_fmlalq_lane_highv4sf): Ditto.
9272         (aarch64_fmlsl_lane_highv4sf): Ditto.
9273         (aarch64_fmlalq_laneq_highv4sf): Ditto.
9274         (aarch64_fmlsl_laneq_highv4sf): Ditto.
9275         * config/aarch64/aarch64-simd.md:
9276         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
9277         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
9278         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
9279         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
9280         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
9281         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
9282         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
9283         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
9284         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
9285         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
9286         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
9287         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
9288         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
9289         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
9290         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
9291         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
9292         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
9293         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
9294         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
9295         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
9296         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
9297         (vfmlsl_low_u32): Ditto.
9298         (vfmlalq_low_u32): Ditto.
9299         (vfmlslq_low_u32): Ditto.
9300         (vfmlal_high_u32): Ditto.
9301         (vfmlsl_high_u32): Ditto.
9302         (vfmlalq_high_u32): Ditto.
9303         (vfmlslq_high_u32): Ditto.
9304         (vfmlal_lane_low_u32): Ditto.
9305         (vfmlsl_lane_low_u32): Ditto.
9306         (vfmlal_laneq_low_u32): Ditto.
9307         (vfmlsl_laneq_low_u32): Ditto.
9308         (vfmlalq_lane_low_u32): Ditto.
9309         (vfmlslq_lane_low_u32): Ditto.
9310         (vfmlalq_laneq_low_u32): Ditto.
9311         (vfmlslq_laneq_low_u32): Ditto.
9312         (vfmlal_lane_high_u32): Ditto.
9313         (vfmlsl_lane_high_u32): Ditto.
9314         (vfmlal_laneq_high_u32): Ditto.
9315         (vfmlsl_laneq_high_u32): Ditto.
9316         (vfmlalq_lane_high_u32): Ditto.
9317         (vfmlslq_lane_high_u32): Ditto.
9318         (vfmlalq_laneq_high_u32): Ditto.
9319         (vfmlslq_laneq_high_u32): Ditto.
9320         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
9321         (AARCH64_FL_FOR_ARCH8_4): New.
9322         (AARCH64_ISA_F16FML): New ISA flag.
9323         (TARGET_F16FML): New feature flag for fp16fml.
9324         (doc/invoke.texi): Document new fp16fml option.
9325
9326 2018-01-10  Michael Collison  <michael.collison@arm.com>
9327
9328         * config/aarch64/aarch64-builtins.c:
9329         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
9330         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9331         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
9332         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
9333         (AARCH64_ISA_SHA3): New ISA flag.
9334         (TARGET_SHA3): New feature flag for sha3.
9335         * config/aarch64/iterators.md (sha512_op): New int attribute.
9336         (CRYPTO_SHA512): New int iterator.
9337         (UNSPEC_SHA512H): New unspec.
9338         (UNSPEC_SHA512H2): Ditto.
9339         (UNSPEC_SHA512SU0): Ditto.
9340         (UNSPEC_SHA512SU1): Ditto.
9341         * config/aarch64/aarch64-simd-builtins.def
9342         (aarch64_crypto_sha512hqv2di): New builtin.
9343         (aarch64_crypto_sha512h2qv2di): Ditto.
9344         (aarch64_crypto_sha512su0qv2di): Ditto.
9345         (aarch64_crypto_sha512su1qv2di): Ditto.
9346         (aarch64_eor3qv8hi): Ditto.
9347         (aarch64_rax1qv2di): Ditto.
9348         (aarch64_xarqv2di): Ditto.
9349         (aarch64_bcaxqv8hi): Ditto.
9350         * config/aarch64/aarch64-simd.md:
9351         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
9352         (aarch64_crypto_sha512su0qv2di): Ditto.
9353         (aarch64_crypto_sha512su1qv2di): Ditto.
9354         (aarch64_eor3qv8hi): Ditto.
9355         (aarch64_rax1qv2di): Ditto.
9356         (aarch64_xarqv2di): Ditto.
9357         (aarch64_bcaxqv8hi): Ditto.
9358         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
9359         (vsha512h2q_u64): Ditto.
9360         (vsha512su0q_u64): Ditto.
9361         (vsha512su1q_u64): Ditto.
9362         (veor3q_u16): Ditto.
9363         (vrax1q_u64): Ditto.
9364         (vxarq_u64): Ditto.
9365         (vbcaxq_u16): Ditto.
9366         * config/arm/types.md (crypto_sha512): New type attribute.
9367         (crypto_sha3): Ditto.
9368         (doc/invoke.texi): Document new sha3 option.
9369
9370 2018-01-10  Michael Collison  <michael.collison@arm.com>
9371
9372         * config/aarch64/aarch64-builtins.c:
9373         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
9374         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9375         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
9376         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
9377         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
9378         (AARCH64_ISA_SM4): New ISA flag.
9379         (TARGET_SM4): New feature flag for sm4.
9380         * config/aarch64/aarch64-simd-builtins.def
9381         (aarch64_sm3ss1qv4si): Ditto.
9382         (aarch64_sm3tt1aq4si): Ditto.
9383         (aarch64_sm3tt1bq4si): Ditto.
9384         (aarch64_sm3tt2aq4si): Ditto.
9385         (aarch64_sm3tt2bq4si): Ditto.
9386         (aarch64_sm3partw1qv4si): Ditto.
9387         (aarch64_sm3partw2qv4si): Ditto.
9388         (aarch64_sm4eqv4si): Ditto.
9389         (aarch64_sm4ekeyqv4si): Ditto.
9390         * config/aarch64/aarch64-simd.md:
9391         (aarch64_sm3ss1qv4si): Ditto.
9392         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
9393         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
9394         (aarch64_sm4eqv4si): Ditto.
9395         (aarch64_sm4ekeyqv4si): Ditto.
9396         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
9397         (sm3part_op): Ditto.
9398         (CRYPTO_SM3TT): Ditto.
9399         (CRYPTO_SM3PART): Ditto.
9400         (UNSPEC_SM3SS1): New unspec.
9401         (UNSPEC_SM3TT1A): Ditto.
9402         (UNSPEC_SM3TT1B): Ditto.
9403         (UNSPEC_SM3TT2A): Ditto.
9404         (UNSPEC_SM3TT2B): Ditto.
9405         (UNSPEC_SM3PARTW1): Ditto.
9406         (UNSPEC_SM3PARTW2): Ditto.
9407         (UNSPEC_SM4E): Ditto.
9408         (UNSPEC_SM4EKEY): Ditto.
9409         * config/aarch64/constraints.md (Ui2): New constraint.
9410         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
9411         * config/arm/types.md (crypto_sm3): New type attribute.
9412         (crypto_sm4): Ditto.
9413         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
9414         (vsm3tt1aq_u32): Ditto.
9415         (vsm3tt1bq_u32): Ditto.
9416         (vsm3tt2aq_u32): Ditto.
9417         (vsm3tt2bq_u32): Ditto.
9418         (vsm3partw1q_u32): Ditto.
9419         (vsm3partw2q_u32): Ditto.
9420         (vsm4eq_u32): Ditto.
9421         (vsm4ekeyq_u32): Ditto.
9422         (doc/invoke.texi): Document new sm4 option.
9423
9424 2018-01-10  Michael Collison  <michael.collison@arm.com>
9425
9426         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
9427         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
9428         (AARCH64_FL_FOR_ARCH8_4): New.
9429         (AARCH64_FL_V8_4): New flag.
9430         (doc/invoke.texi): Document new armv8.4-a option.
9431
9432 2018-01-10  Michael Collison  <michael.collison@arm.com>
9433
9434         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9435         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
9436         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
9437         * config/aarch64/aarch64-option-extension.def: Add
9438         AARCH64_OPT_EXTENSION of 'sha2'.
9439         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
9440         (crypto): Disable sha2 and aes if crypto disabled.
9441         (crypto): Enable aes and sha2 if enabled.
9442         (simd): Disable sha2 and aes if simd disabled.
9443         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
9444         New flags.
9445         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
9446         (TARGET_SHA2): New feature flag for sha2.
9447         (TARGET_AES): New feature flag for aes.
9448         * config/aarch64/aarch64-simd.md:
9449         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
9450         conditional on TARGET_AES.
9451         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
9452         (aarch64_crypto_sha1hsi): Make pattern conditional
9453         on TARGET_SHA2.
9454         (aarch64_crypto_sha1hv4si): Ditto.
9455         (aarch64_be_crypto_sha1hv4si): Ditto.
9456         (aarch64_crypto_sha1su1v4si): Ditto.
9457         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
9458         (aarch64_crypto_sha1su0v4si): Ditto.
9459         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
9460         (aarch64_crypto_sha256su0v4si): Ditto.
9461         (aarch64_crypto_sha256su1v4si): Ditto.
9462         (doc/invoke.texi): Document new aes and sha2 options.
9463
9464 2018-01-10  Martin Sebor  <msebor@redhat.com>
9465
9466         PR tree-optimization/83781
9467         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
9468         as string arrays.
9469
9470 2018-01-11  Martin Sebor  <msebor@gmail.com>
9471             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9472
9473         PR tree-optimization/83501
9474         PR tree-optimization/81703
9475
9476         * tree-ssa-strlen.c (get_string_cst): Rename...
9477         (get_string_len): ...to this.  Handle global constants.
9478         (handle_char_store): Adjust.
9479
9480 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
9481             Jim Wilson  <jimw@sifive.com>
9482
9483         * config/riscv/riscv-protos.h (riscv_output_return): New.
9484         * config/riscv/riscv.c (struct machine_function): New naked_p field.
9485         (riscv_attribute_table, riscv_output_return),
9486         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
9487         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
9488         (riscv_compute_frame_info): Only compute frame->mask if not a naked
9489         function.
9490         (riscv_expand_prologue): Add early return for naked function.
9491         (riscv_expand_epilogue): Likewise.
9492         (riscv_function_ok_for_sibcall): Return false for naked function.
9493         (riscv_set_current_function): New.
9494         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
9495         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
9496         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
9497         * doc/extend.texi (RISC-V Function Attributes): New.
9498
9499 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
9500
9501         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
9502         check for 128-bit long double before checking TCmode.
9503         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
9504         128-bit long doubles before checking TFmode or TCmode.
9505         (FLOAT128_IBM_P): Likewise.
9506
9507 2018-01-10  Martin Sebor  <msebor@redhat.com>
9508
9509         PR tree-optimization/83671
9510         * builtins.c (c_strlen): Unconditionally return zero for the empty
9511         string.
9512         Use -Warray-bounds for warnings.
9513         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
9514         for non-constant array indices with COMPONENT_REF, arrays of
9515         arrays, and pointers to arrays.
9516         (gimple_fold_builtin_strlen): Determine and set length range for
9517         non-constant character arrays.
9518
9519 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
9520
9521         PR middle-end/81897
9522         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
9523         empty blocks.
9524
9525 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
9526
9527         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
9528
9529 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9530
9531         PR target/83399
9532         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
9533         VECTOR_MEM_ALTIVEC_OR_VSX_P.
9534         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
9535         indexed_or_indirect_operand predicate.
9536         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
9537         (*vsx_le_perm_load_v8hi): Likewise.
9538         (*vsx_le_perm_load_v16qi): Likewise.
9539         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
9540         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
9541         (*vsx_le_perm_store_v8hi): Likewise.
9542         (*vsx_le_perm_store_v16qi): Likewise.
9543         (eight unnamed splitters): Likewise.
9544
9545 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9546
9547         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
9548         * config/rs6000/emmintrin.h: Likewise.
9549         * config/rs6000/mmintrin.h: Likewise.
9550         * config/rs6000/xmmintrin.h: Likewise.
9551
9552 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
9553
9554         PR c++/43486
9555         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
9556         "public_flag".
9557         * tree.c (tree_nop_conversion): Return true for location wrapper
9558         nodes.
9559         (maybe_wrap_with_location): New function.
9560         (selftest::check_strip_nops): New function.
9561         (selftest::test_location_wrappers): New function.
9562         (selftest::tree_c_tests): Call it.
9563         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
9564         (maybe_wrap_with_location): New decl.
9565         (EXPR_LOCATION_WRAPPER_P): New macro.
9566         (location_wrapper_p): New inline function.
9567         (tree_strip_any_location_wrapper): New inline function.
9568
9569 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
9570
9571         PR target/83735
9572         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
9573         stack_realign_offset for the largest alignment of stack slot
9574         actually used.
9575         (ix86_find_max_used_stack_alignment): New function.
9576         (ix86_finalize_stack_frame_flags): Use it.  Set
9577         max_used_stack_alignment if we don't realign stack.
9578         * config/i386/i386.h (machine_function): Add
9579         max_used_stack_alignment.
9580
9581 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
9582
9583         * config/arm/arm.opt (-mbranch-cost): New option.
9584         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
9585         account.
9586
9587 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
9588
9589         PR target/83629
9590         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
9591         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
9592
9593 2018-01-10  Richard Biener  <rguenther@suse.de>
9594
9595         PR debug/83765
9596         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
9597         early out so it also covers the case where we have a non-NULL
9598         origin.
9599
9600 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9601
9602         PR tree-optimization/83753
9603         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
9604         for non-strided grouped accesses if the number of elements is 1.
9605
9606 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9607
9608         PR target/81616
9609         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
9610         * i386.h (TARGET_USE_GATHER): Define.
9611         * x86-tune.def (X86_TUNE_USE_GATHER): New.
9612
9613 2018-01-10  Martin Liska  <mliska@suse.cz>
9614
9615         PR bootstrap/82831
9616         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
9617         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
9618         partitioning.
9619         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
9620         CLEANUP_NO_PARTITIONING is not set.
9621
9622 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9623
9624         * doc/rtl.texi: Remove documentation of (const ...) wrappers
9625         for vectors, as a partial revert of r254296.
9626         * rtl.h (const_vec_p): Delete.
9627         (const_vec_duplicate_p): Don't test for vector CONSTs.
9628         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
9629         * expmed.c (make_tree): Likewise.
9630
9631         Revert:
9632         * common.md (E, F): Use CONSTANT_P instead of checking for
9633         CONST_VECTOR.
9634         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
9635         checking for CONST_VECTOR.
9636
9637 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9638
9639         PR middle-end/83575
9640         * predict.c (force_edge_cold): Handle in more sane way edges
9641         with no prediction.
9642
9643 2018-01-09  Carl Love  <cel@us.ibm.com>
9644
9645         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
9646         V4SI, V4SF types.
9647         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
9648         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
9649         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
9650         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
9651         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
9652         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
9653         * config/rs6000/rs6000-protos.h: Add extern defition for
9654         rs6000_generate_float2_double_code.
9655         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
9656         function.
9657         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
9658         (float2_v2df): Add define_expand.
9659
9660 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
9661
9662         PR target/83628
9663         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
9664         op_mode in the force_to_mode call.
9665
9666 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9667
9668         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
9669         instead of checking each element individually.
9670         (aarch64_evpc_uzp): Likewise.
9671         (aarch64_evpc_zip): Likewise.
9672         (aarch64_evpc_ext): Likewise.
9673         (aarch64_evpc_rev): Likewise.
9674         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
9675         instead of checking each element individually.  Return true without
9676         generating rtl if
9677         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
9678         whether all selected elements come from the same input, instead of
9679         checking each element individually.  Remove calls to gen_rtx_REG,
9680         start_sequence and end_sequence and instead assert that no rtl is
9681         generated.
9682
9683 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9684
9685         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
9686         order of HIGH and CONST checks.
9687
9688 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9689
9690         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
9691         if the destination isn't an SSA_NAME.
9692
9693 2018-01-09  Richard Biener  <rguenther@suse.de>
9694
9695         PR tree-optimization/83668
9696         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
9697         move prologue...
9698         (canonicalize_loop_form): ... here, renamed from ...
9699         (canonicalize_loop_closed_ssa_form): ... this and amended to
9700         swap successor edges for loop exit blocks to make us use
9701         the RPO order we need for initial schedule generation.
9702
9703 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
9704
9705         PR tree-optimization/64811
9706         * match.pd: When optimizing comparisons with Inf, avoid
9707         introducing or losing exceptions from comparisons with NaN.
9708
9709 2018-01-09  Martin Liska  <mliska@suse.cz>
9710
9711         PR sanitizer/82517
9712         * asan.c (shadow_mem_size): Add gcc_assert.
9713
9714 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
9715
9716         Don't save registers in main().
9717
9718         PR target/83738
9719         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
9720         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
9721         * config/avr/avr.c (avr_set_current_function): Don't error if
9722         naked, OS_task or OS_main are specified at the same time.
9723         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
9724         OS_main.
9725         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
9726         attribute.
9727         * common/config/avr/avr-common.c (avr_option_optimization_table):
9728         Switch on -mmain-is-OS_task for optimizing compilations.
9729
9730 2018-01-09  Richard Biener  <rguenther@suse.de>
9731
9732         PR tree-optimization/83572
9733         * graphite.c: Include cfganal.h.
9734         (graphite_transform_loops): Connect infinite loops to exit
9735         and remove fake edges at the end.
9736
9737 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9738
9739         * ipa-inline.c (edge_badness): Revert accidental checkin.
9740
9741 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9742
9743         PR ipa/80763
9744         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
9745         symbols; not inline clones.
9746
9747 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
9748
9749         PR target/83507
9750         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
9751         hard registers.  Formatting fixes.
9752
9753         PR preprocessor/83722
9754         * gcc.c (try_generate_repro): Pass
9755         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
9756         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
9757         do_report_bug.
9758
9759 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
9760             Kito Cheng  <kito.cheng@gmail.com>
9761
9762         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
9763         (riscv_leaf_function_p): Delete.
9764         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
9765
9766 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9767
9768         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
9769         function.
9770         (do_ifelse): New function.
9771         (do_isel): New function.
9772         (do_sub3): New function.
9773         (do_add3): New function.
9774         (do_load_mask_compare): New function.
9775         (do_overlap_load_compare): New function.
9776         (expand_compare_loop): New function.
9777         (expand_block_compare): Call expand_compare_loop() when appropriate.
9778         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
9779         option description.
9780         (-mblock-compare-inline-loop-limit): New option.
9781
9782 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9783
9784         PR target/83677
9785         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
9786         Reverse order of second and third operands in first alternative.
9787         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
9788         of first and second elements in UNSPEC_VPERMR vector.
9789         (altivec_expand_vec_perm_le): Likewise.
9790
9791 2018-01-08  Jeff Law  <law@redhat.com>
9792
9793         PR rtl-optimizatin/81308
9794         * tree-switch-conversion.c (cfg_altered): New file scoped static.
9795         (process_switch): If group_case_labels makes a change, then set
9796         cfg_altered.
9797         (pass_convert_switch::execute): If a switch is converted, then
9798         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
9799
9800         PR rtl-optimization/81308
9801         * recog.c (split_all_insns): Conditionally cleanup the CFG after
9802         splitting insns.
9803
9804 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
9805
9806         PR target/83663 - Revert r255946
9807         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
9808         generation for cases where splatting a value is not useful.
9809         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
9810         across a vec_duplicate and a paradoxical subreg forming a vector
9811         mode to a vec_concat.
9812
9813 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9814
9815         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
9816         -march=armv8.3-a variants.
9817         * config/arm/t-multilib: Likewise.
9818         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
9819
9820 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9821
9822         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
9823         to generate rtl.
9824         (cceq_ior_compare_complement): Give it a name so I can use it, and
9825         change boolean_or_operator predicate to boolean_operator so it can
9826         be used to generate a crand.
9827         (eqne): New code iterator.
9828         (bd/bd_neg): New code_attrs.
9829         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
9830         a single define_insn.
9831         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
9832         decrement (bdnzt/bdnzf/bdzt/bdzf).
9833         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
9834         with the new names of the branch decrement patterns, and added the
9835         names of the branch decrement conditional patterns.
9836
9837 2018-01-08  Richard Biener  <rguenther@suse.de>
9838
9839         PR tree-optimization/83563
9840         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
9841         cache.
9842
9843 2018-01-08  Richard Biener  <rguenther@suse.de>
9844
9845         PR middle-end/83713
9846         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
9847
9848 2018-01-08  Richard Biener  <rguenther@suse.de>
9849
9850         PR tree-optimization/83685
9851         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
9852         references to abnormals.
9853
9854 2018-01-08  Richard Biener  <rguenther@suse.de>
9855
9856         PR lto/83719
9857         * dwarf2out.c (output_indirect_strings): Handle empty
9858         skeleton_debug_str_hash.
9859         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
9860
9861 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9862
9863         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
9864         (emit_store_direct): Likewise.
9865         (arc_trampoline_adjust_address): Likewise.
9866         (arc_asm_trampoline_template): New function.
9867         (arc_initialize_trampoline): Use asm_trampoline_template.
9868         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
9869         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
9870         * config/arc/arc.md (flush_icache): Delete pattern.
9871
9872 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9873
9874         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
9875         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
9876         munaligned-access.
9877
9878 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9879
9880         PR target/83681
9881         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
9882         by not USED_FOR_TARGET.
9883         (make_pass_resolve_sw_modes): Likewise.
9884
9885 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9886
9887         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
9888         USED_FOR_TARGET.
9889
9890 2018-01-08  Richard Biener  <rguenther@suse.de>
9891
9892         PR middle-end/83580
9893         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
9894
9895 2018-01-08  Richard Biener  <rguenther@suse.de>
9896
9897         PR middle-end/83517
9898         * match.pd ((t * 2) / 2) -> t): Add missing :c.
9899
9900 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
9901
9902         PR middle-end/81897
9903         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
9904         basic blocks with a small number of successors.
9905         (convert_control_dep_chain_into_preds): Improve handling of
9906         forwarder blocks.
9907         (dump_predicates): Split apart into...
9908         (dump_pred_chain): ...here...
9909         (dump_pred_info): ...and here.
9910         (can_one_predicate_be_invalidated_p): Add debugging printfs.
9911         (can_chain_union_be_invalidated_p): Improve check for invalidation
9912         of paths.
9913         (uninit_uses_cannot_happen): Avoid unnecessary if
9914         convert_control_dep_chain_into_preds yielded nothing.
9915
9916 2018-01-06  Martin Sebor  <msebor@redhat.com>
9917
9918         PR tree-optimization/83640
9919         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
9920         subtracting negative offset from size.
9921         (builtin_access::overlap): Adjust offset bounds of the access to fall
9922         within the size of the object if possible.
9923
9924 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
9925
9926         PR rtl-optimization/83699
9927         * expmed.c (extract_bit_field_1): Restrict the vector usage of
9928         extract_bit_field_as_subreg to cases in which the extracted
9929         value is also a vector.
9930
9931         * lra-constraints.c (process_alt_operands): Test for the equivalence
9932         substitutions when detecting a possible reload cycle.
9933
9934 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
9935
9936         PR debug/83480
9937         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
9938         by default if flag_selective_schedling{,2}.  Formatting fixes.
9939
9940         PR rtl-optimization/83682
9941         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
9942         if it has non-VECTOR_MODE element mode.
9943         (vec_duplicate_p): Likewise.
9944
9945         PR middle-end/83694
9946         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
9947         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
9948
9949 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
9950
9951         PR target/83604
9952         * config/i386/i386-builtin.def
9953         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
9954         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
9955         Require also OPTION_MASK_ISA_AVX512F in addition to
9956         OPTION_MASK_ISA_GFNI.
9957         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
9958         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
9959         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
9960         to OPTION_MASK_ISA_GFNI.
9961         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
9962         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
9963         OPTION_MASK_ISA_AVX512BW.
9964         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
9965         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
9966         addition to OPTION_MASK_ISA_GFNI.
9967         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
9968         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
9969         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
9970         to OPTION_MASK_ISA_GFNI.
9971         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
9972         a requirement for all ISAs rather than any of them with a few
9973         exceptions.
9974         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
9975         processing.
9976         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
9977         bitmasks to be enabled with 3 exceptions, instead of requiring any
9978         enabled ISA with lots of exceptions.
9979         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
9980         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
9981         Change avx512bw in isa attribute to avx512f.
9982         * config/i386/sgxintrin.h: Add license boilerplate.
9983         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
9984         to __AVX512F__ and __AVX512VL to __AVX512VL__.
9985         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
9986         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
9987         defined.
9988         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
9989         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
9990         temporarily sse2 rather than sse if not enabled already.
9991
9992         PR target/83604
9993         * config/i386/sse.md (VI248_VLBW): Rename to ...
9994         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
9995         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
9996         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
9997         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
9998         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
9999         mode iterator instead of VI248_VLBW.
10000
10001 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
10002
10003         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
10004         (record_modified): Skip clobbers; add debug output.
10005         (param_change_prob): Use sreal frequencies.
10006
10007 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
10008
10009         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
10010         punt for user-aligned variables.
10011
10012 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
10013
10014         * tree-chrec.c (chrec_contains_symbols): Return true for
10015         POLY_INT_CST.
10016
10017 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
10018
10019         PR target/82439
10020         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
10021         of (x|y) == x for BICS pattern.
10022
10023 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
10024
10025         PR tree-optimization/83605
10026         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
10027         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
10028         can throw.
10029
10030 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
10031
10032         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
10033         * config/epiphany/rtems.h: New file.
10034
10035 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10036             Uros Bizjak  <ubizjak@gmail.com>
10037
10038         PR target/83554
10039         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
10040         QIreg_operand instead of register_operand predicate.
10041         * config/i386/i386.c (ix86_rop_should_change_byte_p,
10042         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
10043         comments instead of -fmitigate[-_]rop.
10044
10045 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
10046
10047         PR bootstrap/81926
10048         * cgraphunit.c (symbol_table::compile): Switch to text_section
10049         before calling assembly_start debug hook.
10050         * run-rtl-passes.c (run_rtl_passes): Likewise.
10051         Include output.h.
10052
10053 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10054
10055         * tree-vrp.c (extract_range_from_binary_expr_1): Check
10056         range_int_cst_p rather than !symbolic_range_p before calling
10057         extract_range_from_multiplicative_op_1.
10058
10059 2018-01-04  Jeff Law  <law@redhat.com>
10060
10061         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
10062         redundant test in assertion.
10063
10064 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10065
10066         * doc/rtl.texi: Document machine_mode wrapper classes.
10067
10068 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10069
10070         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
10071         using tree_to_uhwi.
10072
10073 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10074
10075         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
10076         the VEC_PERM_EXPR fold to fail.
10077
10078 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10079
10080         PR debug/83585
10081         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
10082         to switched_sections.
10083
10084 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10085
10086         PR target/83680
10087         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
10088         test for d.testing.
10089
10090 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
10091
10092         PR target/83387
10093         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
10094         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
10095
10096 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10097
10098         PR debug/83666
10099         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
10100         is BLKmode and bitpos not zero or mode change is needed.
10101
10102 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10103
10104         PR target/83675
10105         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
10106         TARGET_VIS2.
10107
10108 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
10109
10110         PR target/83628
10111         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
10112         instead of MULT rtx.  Update all corresponding splitters.
10113         (*saddl_se): Ditto.
10114         (*ssub<modesuffix>): Ditto.
10115         (*ssubl_se): Ditto.
10116         (*cmp_sadd_di): Update split patterns.
10117         (*cmp_sadd_si): Ditto.
10118         (*cmp_sadd_sidi): Ditto.
10119         (*cmp_ssub_di): Ditto.
10120         (*cmp_ssub_si): Ditto.
10121         (*cmp_ssub_sidi): Ditto.
10122         * config/alpha/predicates.md (const23_operand): New predicate.
10123         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
10124         Look for ASHIFT, not MULT inner operand.
10125         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
10126
10127 2018-01-04  Martin Liska  <mliska@suse.cz>
10128
10129         PR gcov-profile/83669
10130         * gcov.c (output_intermediate_file): Add version to intermediate
10131         gcov file.
10132         * doc/gcov.texi: Document new field 'version' in intermediate
10133         file format. Fix location of '-k' option of gcov command.
10134
10135 2018-01-04  Martin Liska  <mliska@suse.cz>
10136
10137         PR ipa/82352
10138         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
10139
10140 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10141
10142         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
10143
10144 2018-01-03  Martin Sebor  <msebor@redhat.com>
10145
10146         PR tree-optimization/83655
10147         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
10148         checking calls with invalid arguments.
10149
10150 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10151
10152         * tree-vect-stmts.c (vect_get_store_rhs): New function.
10153         (vectorizable_mask_load_store): Delete.
10154         (vectorizable_call): Return false for masked loads and stores.
10155         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
10156         instead of gimple_assign_rhs1.
10157         (vectorizable_load): Handle IFN_MASK_LOAD.
10158         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
10159
10160 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10161
10162         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
10163         split out from..,
10164         (vectorizable_mask_load_store): ...here.
10165         (vectorizable_load): ...and here.
10166
10167 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10168
10169         * tree-vect-stmts.c (vect_build_all_ones_mask)
10170         (vect_build_zero_merge_argument): New functions, split out from...
10171         (vectorizable_load): ...here.
10172
10173 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10174
10175         * tree-vect-stmts.c (vect_check_store_rhs): New function,
10176         split out from...
10177         (vectorizable_mask_load_store): ...here.
10178         (vectorizable_store): ...and here.
10179
10180 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10181
10182         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
10183         split out from...
10184         (vectorizable_mask_load_store): ...here.
10185
10186 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10187
10188         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
10189         (vect_model_store_cost): Take a vec_load_store_type instead of a
10190         vect_def_type.
10191         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
10192         (vect_model_store_cost): Take a vec_load_store_type instead of a
10193         vect_def_type.
10194         (vectorizable_mask_load_store): Update accordingly.
10195         (vectorizable_store): Likewise.
10196         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
10197
10198 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10199
10200         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
10201         IFN_MASK_LOAD calls here rather than...
10202         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
10203
10204 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10205             Alan Hayward  <alan.hayward@arm.com>
10206             David Sherwood  <david.sherwood@arm.com>
10207
10208         * expmed.c (extract_bit_field_1): For vector extracts,
10209         fall back to extract_bit_field_as_subreg if vec_extract
10210         isn't available.
10211
10212 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10213             Alan Hayward  <alan.hayward@arm.com>
10214             David Sherwood  <david.sherwood@arm.com>
10215
10216         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
10217         they are variable or constant sized.
10218         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
10219         slots for constant-sized data.
10220
10221 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10222             Alan Hayward  <alan.hayward@arm.com>
10223             David Sherwood  <david.sherwood@arm.com>
10224
10225         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
10226         handling COND_EXPRs with boolean comparisons, try to find a better
10227         basis for the mask type than the boolean itself.
10228
10229 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10230
10231         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
10232         is calculated and how it can be overridden.
10233         * genmodes.c (max_bitsize_mode_any_mode): New variable.
10234         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
10235         if defined.
10236         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
10237         if nonzero.
10238
10239 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10240             Alan Hayward  <alan.hayward@arm.com>
10241             David Sherwood  <david.sherwood@arm.com>
10242
10243         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
10244         Remove the mode argument.
10245         (aarch64_simd_valid_immediate): Remove the mode and inverse
10246         arguments.
10247         * config/aarch64/iterators.md (bitsize): New iterator.
10248         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
10249         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
10250         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
10251         aarch64_simd_valid_immediate.
10252         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
10253         (aarch64_reg_or_bic_imm): Likewise.
10254         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
10255         with an insn_type enum and msl with a modifier_type enum.
10256         Replace element_width with a scalar_mode.  Change the shift
10257         to unsigned int.  Add constructors for scalar_float_mode and
10258         scalar_int_mode elements.
10259         (aarch64_vect_float_const_representable_p): Delete.
10260         (aarch64_can_const_movi_rtx_p)
10261         (aarch64_simd_scalar_immediate_valid_for_move)
10262         (aarch64_simd_make_constant): Update call to
10263         aarch64_simd_valid_immediate.
10264         (aarch64_advsimd_valid_immediate_hs): New function.
10265         (aarch64_advsimd_valid_immediate): Likewise.
10266         (aarch64_simd_valid_immediate): Remove mode and inverse
10267         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
10268         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
10269         and aarch64_float_const_representable_p on the result.
10270         (aarch64_output_simd_mov_immediate): Remove mode argument.
10271         Update call to aarch64_simd_valid_immediate and use of
10272         simd_immediate_info.
10273         (aarch64_output_scalar_simd_mov_immediate): Update call
10274         accordingly.
10275
10276 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10277             Alan Hayward  <alan.hayward@arm.com>
10278             David Sherwood  <david.sherwood@arm.com>
10279
10280         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
10281         (mode_nunits): Likewise CONST_MODE_NUNITS.
10282         * machmode.def (ADJUST_NUNITS): Document.
10283         * genmodes.c (mode_data::need_nunits_adj): New field.
10284         (blank_mode): Update accordingly.
10285         (adj_nunits): New variable.
10286         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
10287         parameter.
10288         (emit_mode_size_inline): Set need_bytesize_adj for all modes
10289         listed in adj_nunits.
10290         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
10291         listed in adj_nunits.  Don't emit case statements for such modes.
10292         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
10293         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
10294         nothing if adj_nunits is nonnull.
10295         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
10296         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
10297         (emit_mode_fbit): Update use of print_maybe_const_decl.
10298         (emit_move_size): Likewise.  Treat the array as non-const
10299         if adj_nunits.
10300         (emit_mode_adjustments): Handle adj_nunits.
10301
10302 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10303
10304         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
10305         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
10306         (VECTOR_MODES): Use it.
10307         (make_vector_modes): Take the prefix as an argument.
10308
10309 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10310             Alan Hayward  <alan.hayward@arm.com>
10311             David Sherwood  <david.sherwood@arm.com>
10312
10313         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
10314         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
10315         for MODE_VECTOR_BOOL.
10316         * machmode.def (VECTOR_BOOL_MODE): Document.
10317         * genmodes.c (VECTOR_BOOL_MODE): New macro.
10318         (make_vector_bool_mode): New function.
10319         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
10320         MODE_VECTOR_BOOL.
10321         * lto-streamer-in.c (lto_input_mode_table): Likewise.
10322         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
10323         Likewise.
10324         * stor-layout.c (int_mode_for_mode): Likewise.
10325         * tree.c (build_vector_type_for_mode): Likewise.
10326         * varasm.c (output_constant_pool_2): Likewise.
10327         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
10328         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
10329         for MODE_VECTOR_BOOL.
10330         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
10331         of mode class checks.
10332         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
10333         instead of a list of mode class checks.
10334         (expand_vector_scalar_condition): Likewise.
10335         (type_for_widest_vector_mode): Handle BImode as an inner mode.
10336
10337 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10338             Alan Hayward  <alan.hayward@arm.com>
10339             David Sherwood  <david.sherwood@arm.com>
10340
10341         * machmode.h (mode_size): Change from unsigned short to
10342         poly_uint16_pod.
10343         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
10344         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10345         or if measurement_type is not polynomial.
10346         (fixed_size_mode::includes_p): Check for constant-sized modes.
10347         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
10348         return a poly_uint16 rather than an unsigned short.
10349         (emit_mode_size): Change the type of mode_size from unsigned short
10350         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
10351         (emit_mode_adjustments): Cope with polynomial vector sizes.
10352         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10353         for GET_MODE_SIZE.
10354         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10355         for GET_MODE_SIZE.
10356         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
10357         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
10358         * caller-save.c (setup_save_areas): Likewise.
10359         (replace_reg_with_saved_mem): Likewise.
10360         * calls.c (emit_library_call_value_1): Likewise.
10361         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
10362         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
10363         (gen_lowpart_for_combine): Likewise.
10364         * convert.c (convert_to_integer_1): Likewise.
10365         * cse.c (equiv_constant, cse_insn): Likewise.
10366         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
10367         (cselib_subst_to_values): Likewise.
10368         * dce.c (word_dce_process_block): Likewise.
10369         * df-problems.c (df_word_lr_mark_ref): Likewise.
10370         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
10371         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
10372         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
10373         (rtl_for_decl_location): Likewise.
10374         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
10375         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
10376         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
10377         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
10378         (expand_expr_real_1): Likewise.
10379         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
10380         (pad_below): Likewise.
10381         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10382         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
10383         * ira.c (get_subreg_tracking_sizes): Likewise.
10384         * ira-build.c (ira_create_allocno_objects): Likewise.
10385         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
10386         (ira_sort_regnos_for_alter_reg): Likewise.
10387         * ira-costs.c (record_operand_costs): Likewise.
10388         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
10389         (resolve_simple_move): Likewise.
10390         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
10391         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
10392         (lra_constraints): Likewise.
10393         (CONST_POOL_OK_P): Reject variable-sized modes.
10394         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
10395         (add_pseudo_to_slot, lra_spill): Likewise.
10396         * omp-low.c (omp_clause_aligned_alignment): Likewise.
10397         * optabs-query.c (get_best_extraction_insn): Likewise.
10398         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10399         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
10400         (expand_mult_highpart, valid_multiword_target_p): Likewise.
10401         * recog.c (offsettable_address_addr_space_p): Likewise.
10402         * regcprop.c (maybe_mode_change): Likewise.
10403         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
10404         * regrename.c (build_def_use): Likewise.
10405         * regstat.c (dump_reg_info): Likewise.
10406         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
10407         (find_reloads, find_reloads_subreg_address): Likewise.
10408         * reload1.c (eliminate_regs_1): Likewise.
10409         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
10410         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
10411         (simplify_binary_operation_1, simplify_subreg): Likewise.
10412         * targhooks.c (default_function_arg_padding): Likewise.
10413         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
10414         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
10415         (verify_gimple_assign_ternary): Likewise.
10416         * tree-inline.c (estimate_move_cost): Likewise.
10417         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10418         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
10419         (get_address_cost_ainc): Likewise.
10420         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
10421         (vect_supportable_dr_alignment): Likewise.
10422         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10423         (vectorizable_reduction): Likewise.
10424         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
10425         (vectorizable_operation, vectorizable_load): Likewise.
10426         * tree.c (build_same_sized_truth_vector_type): Likewise.
10427         * valtrack.c (cleanup_auto_inc_dec): Likewise.
10428         * var-tracking.c (emit_note_insn_var_location): Likewise.
10429         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
10430         (ADDR_VEC_ALIGN): Likewise.
10431
10432 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10433             Alan Hayward  <alan.hayward@arm.com>
10434             David Sherwood  <david.sherwood@arm.com>
10435
10436         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
10437         unsigned short.
10438         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10439         or if measurement_type is polynomial.
10440         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
10441         * combine.c (make_extraction): Likewise.
10442         * dse.c (find_shift_sequence): Likewise.
10443         * dwarf2out.c (mem_loc_descriptor): Likewise.
10444         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
10445         (extract_bit_field, extract_low_bits): Likewise.
10446         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
10447         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
10448         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
10449         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
10450         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10451         * reload.c (find_reloads): Likewise.
10452         * reload1.c (alter_reg): Likewise.
10453         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
10454         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
10455         * tree-if-conv.c (predicate_mem_writes): Likewise.
10456         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
10457         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
10458         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
10459         * valtrack.c (dead_debug_insert_temp): Likewise.
10460         * varasm.c (mergeable_constant_section): Likewise.
10461         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
10462
10463 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10464             Alan Hayward  <alan.hayward@arm.com>
10465             David Sherwood  <david.sherwood@arm.com>
10466
10467         * expr.c (expand_assignment): Cope with polynomial mode sizes
10468         when assigning to a CONCAT.
10469
10470 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10471             Alan Hayward  <alan.hayward@arm.com>
10472             David Sherwood  <david.sherwood@arm.com>
10473
10474         * machmode.h (mode_precision): Change from unsigned short to
10475         poly_uint16_pod.
10476         (mode_to_precision): Return a poly_uint16 rather than an unsigned
10477         short.
10478         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
10479         or if measurement_type is not polynomial.
10480         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
10481         in which the mode is already known to be a scalar_int_mode.
10482         * genmodes.c (emit_mode_precision): Change the type of mode_precision
10483         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
10484         initializer.
10485         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10486         for GET_MODE_PRECISION.
10487         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10488         for GET_MODE_PRECISION.
10489         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
10490         as polynomial.
10491         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
10492         (expand_field_assignment, make_extraction): Likewise.
10493         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
10494         (get_last_value): Likewise.
10495         * convert.c (convert_to_integer_1): Likewise.
10496         * cse.c (cse_insn): Likewise.
10497         * expr.c (expand_expr_real_1): Likewise.
10498         * lra-constraints.c (simplify_operand_subreg): Likewise.
10499         * optabs-query.c (can_atomic_load_p): Likewise.
10500         * optabs.c (expand_atomic_load): Likewise.
10501         (expand_atomic_store): Likewise.
10502         * ree.c (combine_reaching_defs): Likewise.
10503         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
10504         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
10505         * tree.h (type_has_mode_precision_p): Likewise.
10506         * ubsan.c (instrument_si_overflow): Likewise.
10507
10508 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10509             Alan Hayward  <alan.hayward@arm.com>
10510             David Sherwood  <david.sherwood@arm.com>
10511
10512         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
10513         polynomial numbers of units.
10514         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
10515         (valid_vector_subparts_p): New function.
10516         (build_vector_type): Remove temporary shim and take the number
10517         of units as a poly_uint64 rather than an int.
10518         (build_opaque_vector_type): Take the number of units as a
10519         poly_uint64 rather than an int.
10520         * tree.c (build_vector_from_ctor): Handle polynomial
10521         TYPE_VECTOR_SUBPARTS.
10522         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
10523         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
10524         (build_vector_from_val): If the number of units is variable,
10525         use build_vec_duplicate_cst for constant operands and
10526         VEC_DUPLICATE_EXPR otherwise.
10527         (make_vector_type): Remove temporary is_constant ().
10528         (build_vector_type, build_opaque_vector_type): Take the number of
10529         units as a poly_uint64 rather than an int.
10530         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
10531         VECTOR_CST_NELTS.
10532         * cfgexpand.c (expand_debug_expr): Likewise.
10533         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
10534         (store_constructor, expand_expr_real_1): Likewise.
10535         (const_scalar_mask_from_tree): Likewise.
10536         * fold-const-call.c (fold_const_reduction): Likewise.
10537         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
10538         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
10539         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
10540         (fold_relational_const): Likewise.
10541         (native_interpret_vector): Likewise.  Change the size from an
10542         int to an unsigned int.
10543         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
10544         TYPE_VECTOR_SUBPARTS.
10545         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
10546         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
10547         duplicating a non-constant operand into a variable-length vector.
10548         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
10549         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
10550         * ipa-icf.c (sem_variable::equals): Likewise.
10551         * match.pd: Likewise.
10552         * omp-simd-clone.c (simd_clone_subparts): Likewise.
10553         * print-tree.c (print_node): Likewise.
10554         * stor-layout.c (layout_type): Likewise.
10555         * targhooks.c (default_builtin_vectorization_cost): Likewise.
10556         * tree-cfg.c (verify_gimple_comparison): Likewise.
10557         (verify_gimple_assign_binary): Likewise.
10558         (verify_gimple_assign_ternary): Likewise.
10559         (verify_gimple_assign_single): Likewise.
10560         * tree-pretty-print.c (dump_generic_node): Likewise.
10561         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10562         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
10563         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
10564         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
10565         (vect_shift_permute_load_chain): Likewise.
10566         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
10567         (expand_vector_condition, optimize_vector_constructor): Likewise.
10568         (lower_vec_perm, get_compute_type): Likewise.
10569         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10570         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
10571         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
10572         (vect_recog_mask_conversion_pattern): Likewise.
10573         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
10574         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
10575         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10576         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
10577         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
10578         (vectorizable_shift, vectorizable_operation, vectorizable_store)
10579         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
10580         (supportable_widening_operation): Likewise.
10581         (supportable_narrowing_operation): Likewise.
10582         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
10583         Likewise.
10584         * varasm.c (output_constant): Likewise.
10585
10586 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10587             Alan Hayward  <alan.hayward@arm.com>
10588             David Sherwood  <david.sherwood@arm.com>
10589
10590         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
10591         so that both the length == 3 and length != 3 cases set up their
10592         own permute vectors.  Add comments explaining why we know the
10593         number of elements is constant.
10594         (vect_permute_load_chain): Likewise.
10595
10596 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10597             Alan Hayward  <alan.hayward@arm.com>
10598             David Sherwood  <david.sherwood@arm.com>
10599
10600         * machmode.h (mode_nunits): Change from unsigned char to
10601         poly_uint16_pod.
10602         (ONLY_FIXED_SIZE_MODES): New macro.
10603         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
10604         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
10605         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
10606         New typedefs.
10607         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
10608         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
10609         or if measurement_type is not polynomial.
10610         * genmodes.c (ZERO_COEFFS): New macro.
10611         (emit_mode_nunits_inline): Make mode_nunits_inline return a
10612         poly_uint16.
10613         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
10614         Use ZERO_COEFFS when emitting initializers.
10615         * data-streamer.h (bp_pack_poly_value): New function.
10616         (bp_unpack_poly_value): Likewise.
10617         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10618         for GET_MODE_NUNITS.
10619         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10620         for GET_MODE_NUNITS.
10621         * tree.c (make_vector_type): Remove temporary shim and make
10622         the real function take the number of units as a poly_uint64
10623         rather than an int.
10624         (build_vector_type_for_mode): Handle polynomial nunits.
10625         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
10626         * emit-rtl.c (const_vec_series_p_1): Likewise.
10627         (gen_rtx_CONST_VECTOR): Likewise.
10628         * fold-const.c (test_vec_duplicate_folding): Likewise.
10629         * genrecog.c (validate_pattern): Likewise.
10630         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
10631         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10632         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
10633         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
10634         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
10635         * rtlanal.c (subreg_get_info): Likewise.
10636         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
10637         (vect_grouped_load_supported): Likewise.
10638         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
10639         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
10640         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
10641         (simplify_const_unary_operation, simplify_binary_operation_1)
10642         (simplify_const_binary_operation, simplify_ternary_operation)
10643         (test_vector_ops_duplicate, test_vector_ops): Likewise.
10644         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
10645         instead of CONST_VECTOR_NUNITS.
10646         * varasm.c (output_constant_pool_2): Likewise.
10647         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
10648         explicit-encoded elements in the XVEC for variable-length vectors.
10649
10650 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10651
10652         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
10653
10654 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10655             Alan Hayward  <alan.hayward@arm.com>
10656             David Sherwood  <david.sherwood@arm.com>
10657
10658         * coretypes.h (fixed_size_mode): Declare.
10659         (fixed_size_mode_pod): New typedef.
10660         * builtins.h (target_builtins::x_apply_args_mode)
10661         (target_builtins::x_apply_result_mode): Change type to
10662         fixed_size_mode_pod.
10663         * builtins.c (apply_args_size, apply_result_size, result_vector)
10664         (expand_builtin_apply_args_1, expand_builtin_apply)
10665         (expand_builtin_return): Update accordingly.
10666
10667 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10668
10669         * cse.c (hash_rtx_cb): Hash only the encoded elements.
10670         * cselib.c (cselib_hash_rtx): Likewise.
10671         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
10672         CONST_VECTOR encoding.
10673
10674 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10675             Jeff Law  <law@redhat.com>
10676
10677         PR target/83641
10678         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
10679         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
10680         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
10681         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
10682
10683         PR target/83641
10684         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
10685         explicitly probe *sp in a noreturn function if there were any callee
10686         register saves or frame pointer is needed.
10687
10688 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10689
10690         PR debug/83621
10691         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
10692         BLKmode for ternary, binary or unary expressions.
10693
10694         PR debug/83645
10695         * var-tracking.c (delete_vta_debug_insn): New inline function.
10696         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
10697         insns from get_insns () to NULL instead of each bb separately.
10698         Use delete_vta_debug_insn.  No longer static.
10699         (vt_debug_insns_local, variable_tracking_main_1): Adjust
10700         delete_vta_debug_insns callers.
10701         * rtl.h (delete_vta_debug_insns): Declare.
10702         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
10703         instead of variable_tracking_main.
10704
10705 2018-01-03  Martin Sebor  <msebor@redhat.com>
10706
10707         PR tree-optimization/83603
10708         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
10709         arguments past the endof the argument list in functions declared
10710         without a prototype.
10711         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
10712         Avoid checking when arguments are null.
10713
10714 2018-01-03  Martin Sebor  <msebor@redhat.com>
10715
10716         PR c/83559
10717         * doc/extend.texi (attribute const): Fix a typo.
10718         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
10719         issuing -Wsuggest-attribute for void functions.
10720
10721 2018-01-03  Martin Sebor  <msebor@redhat.com>
10722
10723         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
10724         offset_int::from instead of wide_int::to_shwi.
10725         (maybe_diag_overlap): Remove assertion.
10726         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
10727         * gimple-ssa-sprintf.c (format_directive): Same.
10728         (parse_directive): Same.
10729         (sprintf_dom_walker::compute_format_length): Same.
10730         (try_substitute_return_value): Same.
10731
10732 2018-01-03  Jeff Law  <law@redhat.com>
10733
10734         PR middle-end/83654
10735         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
10736         non-constant residual for zero at runtime and avoid probing in
10737         that case.  Reorganize code for trailing problem to mirror handling
10738         of the residual.
10739
10740 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10741
10742         PR tree-optimization/83501
10743         * tree-ssa-strlen.c (get_string_cst): New.
10744         (handle_char_store): Call get_string_cst.
10745
10746 2018-01-03  Martin Liska  <mliska@suse.cz>
10747
10748         PR tree-optimization/83593
10749         * tree-ssa-strlen.c: Include tree-cfg.h.
10750         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
10751         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
10752         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
10753         to false.
10754         (strlen_dom_walker::before_dom_children): Call
10755         gimple_purge_dead_eh_edges. Dump tranformation with details
10756         dump flags.
10757         (strlen_dom_walker::before_dom_children): Update call by adding
10758         new argument cleanup_eh.
10759         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
10760
10761 2018-01-03  Martin Liska  <mliska@suse.cz>
10762
10763         PR ipa/83549
10764         * cif-code.def (VARIADIC_THUNK): New enum value.
10765         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
10766         thunks.
10767
10768 2018-01-03  Jan Beulich  <jbeulich@suse.com>
10769
10770         * sse.md (mov<mode>_internal): Tighten condition for when to use
10771         vmovdqu<ssescalarsize> for TI and OI modes.
10772
10773 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10774
10775         Update copyright years.
10776
10777 2018-01-03  Martin Liska  <mliska@suse.cz>
10778
10779         PR ipa/83594
10780         * ipa-visibility.c (function_and_variable_visibility): Skip
10781         functions with noipa attribure.
10782
10783 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10784
10785         * gcc.c (process_command): Update copyright notice dates.
10786         * gcov-dump.c (print_version): Ditto.
10787         * gcov.c (print_version): Ditto.
10788         * gcov-tool.c (print_version): Ditto.
10789         * gengtype.c (create_file): Ditto.
10790         * doc/cpp.texi: Bump @copying's copyright year.
10791         * doc/cppinternals.texi: Ditto.
10792         * doc/gcc.texi: Ditto.
10793         * doc/gccint.texi: Ditto.
10794         * doc/gcov.texi: Ditto.
10795         * doc/install.texi: Ditto.
10796         * doc/invoke.texi: Ditto.
10797
10798 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10799
10800         * vector-builder.h (vector_builder::m_full_nelts): Change from
10801         unsigned int to poly_uint64.
10802         (vector_builder::full_nelts): Update prototype accordingly.
10803         (vector_builder::new_vector): Likewise.
10804         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
10805         (vector_builder::operator ==): Likewise.
10806         (vector_builder::finalize): Likewise.
10807         * int-vector-builder.h (int_vector_builder::int_vector_builder):
10808         Take the number of elements as a poly_uint64 rather than an
10809         unsigned int.
10810         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
10811         from unsigned int to poly_uint64.
10812         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
10813         (vec_perm_indices::new_vector): Likewise.
10814         (vec_perm_indices::length): Likewise.
10815         (vec_perm_indices::nelts_per_input): Likewise.
10816         (vec_perm_indices::input_nelts): Likewise.
10817         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
10818         number of elements per input as a poly_uint64 rather than an
10819         unsigned int.  Use the original encoding for variable-length
10820         vectors, rather than clamping each individual element.
10821         For the second and subsequent elements in each pattern,
10822         clamp the step and base before clamping their sum.
10823         (vec_perm_indices::series_p): Handle polynomial element counts.
10824         (vec_perm_indices::all_in_range_p): Likewise.
10825         (vec_perm_indices_to_tree): Likewise.
10826         (vec_perm_indices_to_rtx): Likewise.
10827         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
10828         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
10829         (tree_vector_builder::new_binary_operation): Handle polynomial
10830         element counts.  Return false if we need to know the number
10831         of elements at compile time.
10832         * fold-const.c (fold_vec_perm): Punt if the number of elements
10833         isn't known at compile time.
10834
10835 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10836
10837         * vec-perm-indices.h (vec_perm_builder): Change element type
10838         from HOST_WIDE_INT to poly_int64.
10839         (vec_perm_indices::element_type): Update accordingly.
10840         (vec_perm_indices::clamp): Handle polynomial element_types.
10841         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
10842         (vec_perm_indices::all_in_range_p): Likewise.
10843         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
10844         than shwi trees.
10845         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
10846         polynomial vec_perm_indices element types.
10847         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
10848         * fold-const.c (fold_vec_perm): Likewise.
10849         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
10850         * tree-vect-generic.c (lower_vec_perm): Likewise.
10851         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
10852         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
10853         element type to HOST_WIDE_INT.
10854
10855 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10856             Alan Hayward  <alan.hayward@arm.com>
10857             David Sherwood  <david.sherwood@arm.com>
10858
10859         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
10860         rather than an int.  Use plus_constant.
10861         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
10862         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
10863
10864 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10865             Alan Hayward  <alan.hayward@arm.com>
10866             David Sherwood  <david.sherwood@arm.com>
10867
10868         * calls.c (emit_call_1, expand_call): Change struct_value_size from
10869         a HOST_WIDE_INT to a poly_int64.
10870
10871 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10872             Alan Hayward  <alan.hayward@arm.com>
10873             David Sherwood  <david.sherwood@arm.com>
10874
10875         * calls.c (load_register_parameters): Cope with polynomial
10876         mode sizes.  Require a constant size for BLKmode parameters
10877         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
10878         forces a parameter to be padded at the lsb end in order to
10879         fill a complete number of words, require the parameter size
10880         to be ordered wrt UNITS_PER_WORD.
10881
10882 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10883             Alan Hayward  <alan.hayward@arm.com>
10884             David Sherwood  <david.sherwood@arm.com>
10885
10886         * reload1.c (spill_stack_slot_width): Change element type
10887         from unsigned int to poly_uint64_pod.
10888         (alter_reg): Treat mode sizes as polynomial.
10889
10890 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10891             Alan Hayward  <alan.hayward@arm.com>
10892             David Sherwood  <david.sherwood@arm.com>
10893
10894         * reload.c (complex_word_subreg_p): New function.
10895         (reload_inner_reg_of_subreg, push_reload): Use it.
10896
10897 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10898             Alan Hayward  <alan.hayward@arm.com>
10899             David Sherwood  <david.sherwood@arm.com>
10900
10901         * lra-constraints.c (process_alt_operands): Reject matched
10902         operands whose sizes aren't ordered.
10903         (match_reload): Refer to this check here.
10904
10905 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10906             Alan Hayward  <alan.hayward@arm.com>
10907             David Sherwood  <david.sherwood@arm.com>
10908
10909         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
10910         that the mode size is in the set {1, 2, 4, 8, 16}.
10911
10912 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10913             Alan Hayward  <alan.hayward@arm.com>
10914             David Sherwood  <david.sherwood@arm.com>
10915
10916         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
10917         Use plus_constant instead of gen_rtx_PLUS.
10918
10919 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10920             Alan Hayward  <alan.hayward@arm.com>
10921             David Sherwood  <david.sherwood@arm.com>
10922
10923         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
10924         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
10925         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
10926         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
10927         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
10928         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
10929         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
10930         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
10931         * config/i386/i386.c (ix86_push_rounding): ...this new function.
10932         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
10933         a poly_int64.
10934         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
10935         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
10936         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
10937         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
10938         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
10939         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
10940         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
10941         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
10942         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
10943         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
10944         function.
10945         * expr.c (emit_move_resolve_push): Treat the input and result
10946         of PUSH_ROUNDING as a poly_int64.
10947         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
10948         (emit_push_insn): Likewise.
10949         * lra-eliminations.c (mark_not_eliminable): Likewise.
10950         * recog.c (push_operand): Likewise.
10951         * reload1.c (elimination_effects): Likewise.
10952         * rtlanal.c (nonzero_bits1): Likewise.
10953         * calls.c (store_one_arg): Likewise.  Require the padding to be
10954         known at compile time.
10955
10956 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10957             Alan Hayward  <alan.hayward@arm.com>
10958             David Sherwood  <david.sherwood@arm.com>
10959
10960         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
10961         Use plus_constant instead of gen_rtx_PLUS.
10962
10963 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10964             Alan Hayward  <alan.hayward@arm.com>
10965             David Sherwood  <david.sherwood@arm.com>
10966
10967         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
10968         rather than an int.
10969
10970 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10971             Alan Hayward  <alan.hayward@arm.com>
10972             David Sherwood  <david.sherwood@arm.com>
10973
10974         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
10975         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
10976         via stack temporaries.  Treat the mode size as polynomial too.
10977
10978 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10979             Alan Hayward  <alan.hayward@arm.com>
10980             David Sherwood  <david.sherwood@arm.com>
10981
10982         * expr.c (expand_expr_real_2): When handling conversions involving
10983         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
10984         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
10985         as a poly_uint64 too.
10986
10987 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10988             Alan Hayward  <alan.hayward@arm.com>
10989             David Sherwood  <david.sherwood@arm.com>
10990
10991         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
10992
10993 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10994             Alan Hayward  <alan.hayward@arm.com>
10995             David Sherwood  <david.sherwood@arm.com>
10996
10997         * combine.c (can_change_dest_mode): Handle polynomial
10998         REGMODE_NATURAL_SIZE.
10999         * expmed.c (store_bit_field_1): Likewise.
11000         * expr.c (store_constructor): Likewise.
11001         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
11002         and polynomial REGMODE_NATURAL_SIZE.
11003         (gen_lowpart_common): Likewise.
11004         * reginfo.c (record_subregs_of_mode): Likewise.
11005         * rtlanal.c (read_modify_subreg_p): Likewise.
11006
11007 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11008             Alan Hayward  <alan.hayward@arm.com>
11009             David Sherwood  <david.sherwood@arm.com>
11010
11011         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
11012         numbers of elements.
11013
11014 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11015             Alan Hayward  <alan.hayward@arm.com>
11016             David Sherwood  <david.sherwood@arm.com>
11017
11018         * match.pd: Cope with polynomial numbers of vector elements.
11019
11020 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11021             Alan Hayward  <alan.hayward@arm.com>
11022             David Sherwood  <david.sherwood@arm.com>
11023
11024         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
11025         in a POINTER_PLUS_EXPR.
11026
11027 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11028             Alan Hayward  <alan.hayward@arm.com>
11029             David Sherwood  <david.sherwood@arm.com>
11030
11031         * omp-simd-clone.c (simd_clone_subparts): New function.
11032         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
11033         (ipa_simd_modify_function_body): Likewise.
11034
11035 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11036             Alan Hayward  <alan.hayward@arm.com>
11037             David Sherwood  <david.sherwood@arm.com>
11038
11039         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
11040         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
11041         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
11042         (expand_vector_condition, vector_element): Likewise.
11043         (subparts_gt): New function.
11044         (get_compute_type): Use subparts_gt.
11045         (count_type_subparts): Delete.
11046         (expand_vector_operations_1): Use subparts_gt instead of
11047         count_type_subparts.
11048
11049 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11050             Alan Hayward  <alan.hayward@arm.com>
11051             David Sherwood  <david.sherwood@arm.com>
11052
11053         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
11054         (vect_compile_time_alias): ...this new function.  Do the calculation
11055         on poly_ints rather than trees.
11056         (vect_prune_runtime_alias_test_list): Update call accordingly.
11057
11058 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11059             Alan Hayward  <alan.hayward@arm.com>
11060             David Sherwood  <david.sherwood@arm.com>
11061
11062         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
11063         numbers of units.
11064         (vect_schedule_slp_instance): Likewise.
11065
11066 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11067             Alan Hayward  <alan.hayward@arm.com>
11068             David Sherwood  <david.sherwood@arm.com>
11069
11070         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
11071         constant and extern definitions for variable-length vectors.
11072         (vect_get_constant_vectors): Note that the number of units
11073         is known to be constant.
11074
11075 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11076             Alan Hayward  <alan.hayward@arm.com>
11077             David Sherwood  <david.sherwood@arm.com>
11078
11079         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
11080         of units as polynomial.  Choose between WIDE and NARROW based
11081         on multiple_p.
11082
11083 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11084             Alan Hayward  <alan.hayward@arm.com>
11085             David Sherwood  <david.sherwood@arm.com>
11086
11087         * tree-vect-stmts.c (simd_clone_subparts): New function.
11088         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
11089
11090 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11091             Alan Hayward  <alan.hayward@arm.com>
11092             David Sherwood  <david.sherwood@arm.com>
11093
11094         * tree-vect-stmts.c (vectorizable_call): Treat the number of
11095         vectors as polynomial.  Use build_index_vector for
11096         IFN_GOMP_SIMD_LANE.
11097
11098 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11099             Alan Hayward  <alan.hayward@arm.com>
11100             David Sherwood  <david.sherwood@arm.com>
11101
11102         * tree-vect-stmts.c (get_load_store_type): Treat the number of
11103         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
11104         for variable-length vectors.
11105         (vectorizable_mask_load_store): Treat the number of units as
11106         polynomial, asserting that it is constant if the condition has
11107         already been enforced.
11108         (vectorizable_store, vectorizable_load): Likewise.
11109
11110 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11111             Alan Hayward  <alan.hayward@arm.com>
11112             David Sherwood  <david.sherwood@arm.com>
11113
11114         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
11115         of units as polynomial.  Punt if we can't tell at compile time
11116         which vector contains the final result.
11117
11118 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11119             Alan Hayward  <alan.hayward@arm.com>
11120             David Sherwood  <david.sherwood@arm.com>
11121
11122         * tree-vect-loop.c (vectorizable_induction): Treat the number
11123         of units as polynomial.  Punt on SLP inductions.  Use an integer
11124         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
11125         cast of such a series for variable-length floating-point
11126         reductions.
11127
11128 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11129             Alan Hayward  <alan.hayward@arm.com>
11130             David Sherwood  <david.sherwood@arm.com>
11131
11132         * tree.h (build_index_vector): Declare.
11133         * tree.c (build_index_vector): New function.
11134         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
11135         of units as polynomial, forcibly converting it to a constant if
11136         vectorizable_reduction has already enforced the condition.
11137         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
11138         to create a {1,2,3,...} vector.
11139         (vectorizable_reduction): Treat the number of units as polynomial.
11140         Choose vectype_in based on the largest scalar element size rather
11141         than the smallest number of units.  Enforce the restrictions
11142         relied on above.
11143
11144 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11145             Alan Hayward  <alan.hayward@arm.com>
11146             David Sherwood  <david.sherwood@arm.com>
11147
11148         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
11149         number of units as polynomial.
11150
11151 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11152             Alan Hayward  <alan.hayward@arm.com>
11153             David Sherwood  <david.sherwood@arm.com>
11154
11155         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
11156         * target.def (autovectorize_vector_sizes): Return the vector sizes
11157         by pointer, using vector_sizes rather than a bitmask.
11158         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
11159         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
11160         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
11161         Likewise.
11162         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
11163         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
11164         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
11165         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
11166         * omp-general.c (omp_max_vf): Likewise.
11167         * omp-low.c (omp_clause_aligned_alignment): Likewise.
11168         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
11169         * tree-vect-loop.c (vect_analyze_loop): Likewise.
11170         * tree-vect-slp.c (vect_slp_bb): Likewise.
11171         * doc/tm.texi: Regenerate.
11172         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
11173         to a poly_uint64.
11174         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
11175         the vector size as a poly_uint64 rather than an unsigned int.
11176         (current_vector_size): Change from an unsigned int to a poly_uint64.
11177         (get_vectype_for_scalar_type): Update accordingly.
11178         * tree.h (build_truth_vector_type): Take the size and number of
11179         units as a poly_uint64 rather than an unsigned int.
11180         (build_vector_type): Add a temporary overload that takes
11181         the number of units as a poly_uint64 rather than an unsigned int.
11182         * tree.c (make_vector_type): Likewise.
11183         (build_truth_vector_type): Take the number of units as a poly_uint64
11184         rather than an unsigned int.
11185
11186 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11187             Alan Hayward  <alan.hayward@arm.com>
11188             David Sherwood  <david.sherwood@arm.com>
11189
11190         * target.def (get_mask_mode): Take the number of units and length
11191         as poly_uint64s rather than unsigned ints.
11192         * targhooks.h (default_get_mask_mode): Update accordingly.
11193         * targhooks.c (default_get_mask_mode): Likewise.
11194         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
11195         * doc/tm.texi: Regenerate.
11196
11197 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11198             Alan Hayward  <alan.hayward@arm.com>
11199             David Sherwood  <david.sherwood@arm.com>
11200
11201         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
11202         * omp-general.c (omp_max_vf): Likewise.
11203         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
11204         (expand_omp_simd): Handle polynomial safelen.
11205         * omp-low.c (omplow_simd_context): Add a default constructor.
11206         (omplow_simd_context::max_vf): Change from int to poly_uint64.
11207         (lower_rec_simd_input_clauses): Update accordingly.
11208         (lower_rec_input_clauses): Likewise.
11209
11210 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11211             Alan Hayward  <alan.hayward@arm.com>
11212             David Sherwood  <david.sherwood@arm.com>
11213
11214         * tree-vectorizer.h (vect_nunits_for_cost): New function.
11215         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
11216         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
11217         (vect_analyze_slp_cost): Likewise.
11218         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
11219         (vect_model_load_cost): Likewise.
11220
11221 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11222             Alan Hayward  <alan.hayward@arm.com>
11223             David Sherwood  <david.sherwood@arm.com>
11224
11225         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
11226         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
11227         from an unsigned int * to a poly_uint64_pod *.
11228         (calculate_unrolling_factor): New function.
11229         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
11230
11231 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11232             Alan Hayward  <alan.hayward@arm.com>
11233             David Sherwood  <david.sherwood@arm.com>
11234
11235         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
11236         from an unsigned int to a poly_uint64.
11237         (_loop_vec_info::slp_unrolling_factor): Likewise.
11238         (_loop_vec_info::vectorization_factor): Change from an int
11239         to a poly_uint64.
11240         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
11241         (vect_get_num_vectors): New function.
11242         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
11243         (vect_get_num_copies): Use vect_get_num_vectors.
11244         (vect_analyze_data_ref_dependences): Change max_vf from an int *
11245         to an unsigned int *.
11246         (vect_analyze_data_refs): Change min_vf from an int * to a
11247         poly_uint64 *.
11248         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
11249         than an unsigned HOST_WIDE_INT.
11250         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
11251         (vect_analyze_data_ref_dependence): Change max_vf from an int *
11252         to an unsigned int *.
11253         (vect_analyze_data_ref_dependences): Likewise.
11254         (vect_compute_data_ref_alignment): Handle polynomial vf.
11255         (vect_enhance_data_refs_alignment): Likewise.
11256         (vect_prune_runtime_alias_test_list): Likewise.
11257         (vect_shift_permute_load_chain): Likewise.
11258         (vect_supportable_dr_alignment): Likewise.
11259         (dependence_distance_ge_vf): Take the vectorization factor as a
11260         poly_uint64 rather than an unsigned HOST_WIDE_INT.
11261         (vect_analyze_data_refs): Change min_vf from an int * to a
11262         poly_uint64 *.
11263         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
11264         vfm1 as a poly_uint64 rather than an int.  Make the same change
11265         for the returned bound_scalar.
11266         (vect_gen_vector_loop_niters): Handle polynomial vf.
11267         (vect_do_peeling): Likewise.  Update call to
11268         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
11269         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
11270         be constant.
11271         * tree-vect-loop.c (vect_determine_vectorization_factor)
11272         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
11273         (vect_get_known_peeling_cost): Likewise.
11274         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
11275         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
11276         (vect_transform_loop): Likewise.  Use the lowest possible VF when
11277         updating the upper bounds of the loop.
11278         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
11279         rather than an int.
11280         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
11281         polynomial unroll factors.
11282         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
11283         (vect_make_slp_decision): Likewise.
11284         (vect_supported_load_permutation_p): Likewise, and polynomial
11285         vf too.
11286         (vect_analyze_slp_cost): Handle polynomial vf.
11287         (vect_slp_analyze_node_operations): Likewise.
11288         (vect_slp_analyze_bb_1): Likewise.
11289         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
11290         than an unsigned HOST_WIDE_INT.
11291         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
11292         (vectorizable_load): Handle polynomial vf.
11293         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
11294         a poly_uint64.
11295         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
11296
11297 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11298             Alan Hayward  <alan.hayward@arm.com>
11299             David Sherwood  <david.sherwood@arm.com>
11300
11301         * match.pd: Handle bit operations involving three constants
11302         and try to fold one pair.
11303
11304 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11305
11306         * tree-vect-loop-manip.c: Include gimple-fold.h.
11307         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
11308         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
11309         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
11310         Add a path that uses a step of VF instead of 1, but disable it
11311         for now.
11312         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
11313         and niters_no_overflow parameters.  Update calls to
11314         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
11315         Create a new SSA name if the latter choses to use a ste other
11316         than zero, and return it via niters_vector_mult_vf_var.
11317         * tree-vect-loop.c (vect_transform_loop): Update calls to
11318         vect_do_peeling, vect_gen_vector_loop_niters and
11319         slpeel_make_loop_iterate_ntimes.
11320         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
11321         (vect_gen_vector_loop_niters): Update declarations after above changes.
11322
11323 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
11324
11325         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
11326         128-bit round to integer instructions.
11327         (ceil<mode>2): Likewise.
11328         (btrunc<mode>2): Likewise.
11329         (round<mode>2): Likewise.
11330
11331 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11332
11333         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
11334         unaligned VSX load/store on P8/P9.
11335         (expand_block_clear): Allow the use of unaligned VSX
11336         load/store on P8/P9.
11337
11338 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
11339
11340         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
11341         New function.
11342         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
11343         swap associated with both a load and a store.
11344
11345 2018-01-02  Andrew Waterman  <andrew@sifive.com>
11346
11347         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
11348         * config/riscv/riscv.md (clear_cache): Use it.
11349
11350 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
11351
11352         * web.c: Remove out-of-date comment.
11353
11354 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11355
11356         * expr.c (fixup_args_size_notes): Check that any existing
11357         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
11358         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
11359         (emit_single_push_insn): ...here.
11360
11361 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11362
11363         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
11364         (const_vector_encoded_nelts): New function.
11365         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
11366         (const_vector_int_elt, const_vector_elt): Declare.
11367         * emit-rtl.c (const_vector_int_elt_1): New function.
11368         (const_vector_elt): Likewise.
11369         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
11370         of CONST_VECTOR_ELT.
11371
11372 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11373
11374         * expr.c: Include rtx-vector-builder.h.
11375         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
11376         directly on the tree encoding.
11377         (const_vector_from_tree): Likewise.
11378         * optabs.c: Include rtx-vector-builder.h.
11379         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
11380         sequence of "u" values.
11381         * vec-perm-indices.c: Include rtx-vector-builder.h.
11382         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
11383         directly on the vec_perm_indices encoding.
11384
11385 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11386
11387         * doc/rtl.texi (const_vector): Describe new encoding scheme.
11388         * Makefile.in (OBJS): Add rtx-vector-builder.o.
11389         * rtx-vector-builder.h: New file.
11390         * rtx-vector-builder.c: Likewise.
11391         * rtl.h (rtx_def::u2): Add a const_vector field.
11392         (CONST_VECTOR_NPATTERNS): New macro.
11393         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
11394         (CONST_VECTOR_DUPLICATE_P): Likewise.
11395         (CONST_VECTOR_STEPPED_P): Likewise.
11396         (CONST_VECTOR_ENCODED_ELT): Likewise.
11397         (const_vec_duplicate_p): Check for a duplicated vector encoding.
11398         (unwrap_const_vec_duplicate): Likewise.
11399         (const_vec_series_p): Check for a non-duplicated vector encoding.
11400         Say that the function only returns true for integer vectors.
11401         * emit-rtl.c: Include rtx-vector-builder.h.
11402         (gen_const_vec_duplicate_1): Delete.
11403         (gen_const_vector): Call gen_const_vec_duplicate instead of
11404         gen_const_vec_duplicate_1.
11405         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
11406         (gen_const_vec_duplicate): Use rtx_vector_builder.
11407         (gen_const_vec_series): Likewise.
11408         (gen_rtx_CONST_VECTOR): Likewise.
11409         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
11410         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11411         Build a new vector rather than modifying a CONST_VECTOR in-place.
11412         (handle_special_swappables): Update call accordingly.
11413         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
11414         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11415         Build a new vector rather than modifying a CONST_VECTOR in-place.
11416         (handle_special_swappables): Update call accordingly.
11417
11418 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11419
11420         * simplify-rtx.c (simplify_const_binary_operation): Use
11421         CONST_VECTOR_ELT instead of XVECEXP.
11422
11423 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11424
11425         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
11426         the selector elements to be different from the data elements
11427         if the selector is a VECTOR_CST.
11428         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
11429         ssizetype for the selector.
11430
11431 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11432
11433         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
11434         before testing each element individually.
11435         * tree-vect-generic.c (lower_vec_perm): Likewise.
11436
11437 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11438
11439         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
11440         * selftest-run-tests.c (selftest::run_tests): Call it.
11441         * vector-builder.h (vector_builder::operator ==): New function.
11442         (vector_builder::operator !=): Likewise.
11443         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
11444         (vec_perm_indices::all_from_input_p): New function.
11445         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
11446         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
11447         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
11448         instead of reading the VECTOR_CST directly.  Detect whether both
11449         vector inputs are the same before constructing the vec_perm_indices,
11450         and update the number of inputs argument accordingly.  Use the
11451         utility functions added above.  Only construct sel2 if we need to.
11452
11453 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11454
11455         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
11456         the broadcast of the low byte.
11457         (expand_mult_highpart): Use an explicit encoding for the permutes.
11458         * optabs-query.c (can_mult_highpart_p): Likewise.
11459         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
11460         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11461         (vectorizable_bswap): Likewise.
11462         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
11463         explicit encoding for the power-of-2 permutes.
11464         (vect_permute_store_chain): Likewise.
11465         (vect_grouped_load_supported): Likewise.
11466         (vect_permute_load_chain): Likewise.
11467
11468 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11469
11470         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
11471         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
11472         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
11473         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
11474         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
11475         (vect_gen_perm_mask_any): Likewise.
11476
11477 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11478
11479         * int-vector-builder.h: New file.
11480         * vec-perm-indices.h: Include int-vector-builder.h.
11481         (vec_perm_indices): Redefine as an int_vector_builder.
11482         (auto_vec_perm_indices): Delete.
11483         (vec_perm_builder): Redefine as a stand-alone class.
11484         (vec_perm_indices::vec_perm_indices): New function.
11485         (vec_perm_indices::clamp): Likewise.
11486         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
11487         (vec_perm_indices::new_vector): New function.
11488         (vec_perm_indices::new_expanded_vector): Update for new
11489         vec_perm_indices class.
11490         (vec_perm_indices::rotate_inputs): New function.
11491         (vec_perm_indices::all_in_range_p): Operate directly on the
11492         encoded form, without computing elided elements.
11493         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
11494         encoding.  Update for new vec_perm_indices class.
11495         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
11496         the given vec_perm_builder.
11497         (expand_vec_perm_var): Update vec_perm_builder constructor.
11498         (expand_mult_highpart): Use vec_perm_builder instead of
11499         auto_vec_perm_indices.
11500         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
11501         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
11502         or double series encoding as appropriate.
11503         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
11504         vec_perm_indices instead of auto_vec_perm_indices.
11505         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11506         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11507         (vect_permute_store_chain): Likewise.
11508         (vect_grouped_load_supported): Likewise.
11509         (vect_permute_load_chain): Likewise.
11510         (vect_shift_permute_load_chain): Likewise.
11511         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11512         (vect_transform_slp_perm_load): Likewise.
11513         (vect_schedule_slp_instance): Likewise.
11514         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11515         (vectorizable_mask_load_store): Likewise.
11516         (vectorizable_bswap): Likewise.
11517         (vectorizable_store): Likewise.
11518         (vectorizable_load): Likewise.
11519         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
11520         vec_perm_indices instead of auto_vec_perm_indices.  Use
11521         tree_to_vec_perm_builder to read the vector from a tree.
11522         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
11523         vec_perm_builder instead of a vec_perm_indices.
11524         (have_whole_vector_shift): Use vec_perm_builder and
11525         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
11526         truncation to calc_vec_perm_mask_for_shift.
11527         (vect_create_epilog_for_reduction): Likewise.
11528         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
11529         from auto_vec_perm_indices to vec_perm_indices.
11530         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11531         instead of changing individual elements.
11532         (aarch64_vectorize_vec_perm_const): Use new_vector to install
11533         the vector in d.perm.
11534         * config/arm/arm.c (expand_vec_perm_d::perm): Change
11535         from auto_vec_perm_indices to vec_perm_indices.
11536         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11537         instead of changing individual elements.
11538         (arm_vectorize_vec_perm_const): Use new_vector to install
11539         the vector in d.perm.
11540         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
11541         Update vec_perm_builder constructor.
11542         (rs6000_expand_interleave): Likewise.
11543         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
11544         (rs6000_expand_interleave): Likewise.
11545
11546 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11547
11548         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
11549         to qimode could truncate the indices.
11550         * optabs.c (expand_vec_perm_var): Likewise.
11551
11552 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11553
11554         * Makefile.in (OBJS): Add vec-perm-indices.o.
11555         * vec-perm-indices.h: New file.
11556         * vec-perm-indices.c: Likewise.
11557         * target.h (vec_perm_indices): Replace with a forward class
11558         declaration.
11559         (auto_vec_perm_indices): Move to vec-perm-indices.h.
11560         * optabs.h: Include vec-perm-indices.h.
11561         (expand_vec_perm): Delete.
11562         (selector_fits_mode_p, expand_vec_perm_var): Declare.
11563         (expand_vec_perm_const): Declare.
11564         * target.def (vec_perm_const_ok): Replace with...
11565         (vec_perm_const): ...this new hook.
11566         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
11567         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
11568         * doc/tm.texi: Regenerate.
11569         * optabs.def (vec_perm_const): Delete.
11570         * doc/md.texi (vec_perm_const): Likewise.
11571         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
11572         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
11573         expand_vec_perm for constant permutation vectors.  Assert that
11574         the mode of variable permutation vectors is the integer equivalent
11575         of the mode that is being permuted.
11576         * optabs-query.h (selector_fits_mode_p): Declare.
11577         * optabs-query.c: Include vec-perm-indices.h.
11578         (selector_fits_mode_p): New function.
11579         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
11580         is defined, instead of checking whether the vec_perm_const_optab
11581         exists.  Use targetm.vectorize.vec_perm_const instead of
11582         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
11583         fit in the vector mode before using a variable permute.
11584         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
11585         vec_perm_indices instead of an rtx.
11586         (expand_vec_perm): Replace with...
11587         (expand_vec_perm_const): ...this new function.  Take the selector
11588         as a vec_perm_indices rather than an rtx.  Also take the mode of
11589         the selector.  Update call to shift_amt_for_vec_perm_mask.
11590         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
11591         Use vec_perm_indices::new_expanded_vector to expand the original
11592         selector into bytes.  Check whether the indices fit in the vector
11593         mode before using a variable permute.
11594         (expand_vec_perm_var): Make global.
11595         (expand_mult_highpart): Use expand_vec_perm_const.
11596         * fold-const.c: Includes vec-perm-indices.h.
11597         * tree-ssa-forwprop.c: Likewise.
11598         * tree-vect-data-refs.c: Likewise.
11599         * tree-vect-generic.c: Likewise.
11600         * tree-vect-loop.c: Likewise.
11601         * tree-vect-slp.c: Likewise.
11602         * tree-vect-stmts.c: Likewise.
11603         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
11604         Delete.
11605         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
11606         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
11607         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
11608         (aarch64_vectorize_vec_perm_const): ...this new function.
11609         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11610         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11611         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
11612         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
11613         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11614         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11615         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
11616         into...
11617         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
11618         check for NEON modes.
11619         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
11620         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
11621         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
11622         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
11623         into...
11624         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
11625         the old VEC_PERM_CONST conditions.
11626         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
11627         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
11628         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
11629         (ia64_vectorize_vec_perm_const_ok): Merge into...
11630         (ia64_vectorize_vec_perm_const): ...this new function.
11631         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
11632         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
11633         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
11634         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
11635         * config/mips/mips.c (mips_expand_vec_perm_const)
11636         (mips_vectorize_vec_perm_const_ok): Merge into...
11637         (mips_vectorize_vec_perm_const): ...this new function.
11638         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
11639         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
11640         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
11641         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
11642         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
11643         (rs6000_expand_vec_perm_const): Delete.
11644         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
11645         Delete.
11646         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11647         (altivec_expand_vec_perm_const_le): Take each operand individually.
11648         Operate on constant selectors rather than rtxes.
11649         (altivec_expand_vec_perm_const): Likewise.  Update call to
11650         altivec_expand_vec_perm_const_le.
11651         (rs6000_expand_vec_perm_const): Delete.
11652         (rs6000_vectorize_vec_perm_const_ok): Delete.
11653         (rs6000_vectorize_vec_perm_const): New function.
11654         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11655         an element count and rtx array.
11656         (rs6000_expand_extract_even): Update call accordingly.
11657         (rs6000_expand_interleave): Likewise.
11658         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
11659         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
11660         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
11661         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
11662         (rs6000_expand_vec_perm_const): Delete.
11663         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11664         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11665         (altivec_expand_vec_perm_const_le): Take each operand individually.
11666         Operate on constant selectors rather than rtxes.
11667         (altivec_expand_vec_perm_const): Likewise.  Update call to
11668         altivec_expand_vec_perm_const_le.
11669         (rs6000_expand_vec_perm_const): Delete.
11670         (rs6000_vectorize_vec_perm_const_ok): Delete.
11671         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
11672         reference to the SPE evmerge intructions.
11673         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11674         an element count and rtx array.
11675         (rs6000_expand_extract_even): Update call accordingly.
11676         (rs6000_expand_interleave): Likewise.
11677         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
11678         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
11679         new function.
11680         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11681
11682 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11683
11684         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
11685         vector mode and that that mode matches the mode of the data
11686         being permuted.
11687         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
11688         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
11689         directly using expand_vec_perm_1 when forcing selectors into
11690         registers.
11691         (expand_vec_perm_var): New function, split out from expand_vec_perm.
11692
11693 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11694
11695         * optabs-query.h (can_vec_perm_p): Delete.
11696         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
11697         * optabs-query.c (can_vec_perm_p): Split into...
11698         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
11699         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
11700         particular selector is valid.
11701         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11702         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11703         (vect_grouped_load_supported): Likewise.
11704         (vect_shift_permute_load_chain): Likewise.
11705         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11706         (vect_transform_slp_perm_load): Likewise.
11707         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11708         (vectorizable_bswap): Likewise.
11709         (vect_gen_perm_mask_checked): Likewise.
11710         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
11711         implementations of variable permutation vectors into account
11712         when deciding which selector to use.
11713         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
11714         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
11715         with a false third argument.
11716         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
11717         to test whether the constant selector is valid and can_vec_perm_var_p
11718         to test whether a variable selector is valid.
11719
11720 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11721
11722         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
11723         * optabs-query.c (can_vec_perm_p): Likewise.
11724         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
11725         instead of vec_perm_indices.
11726         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
11727         (vect_gen_perm_mask_checked): Likewise,
11728         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
11729         (vect_gen_perm_mask_checked): Likewise,
11730
11731 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11732
11733         * optabs-query.h (qimode_for_vec_perm): Declare.
11734         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
11735         (qimode_for_vec_perm): ...this new function.
11736         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
11737
11738 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11739
11740         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
11741         does not have a conditional at the top.
11742
11743 2018-01-02  Richard Biener  <rguenther@suse.de>
11744
11745         * ipa-inline.c (big_speedup_p): Fix expression.
11746
11747 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11748
11749         PR target/81616
11750         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
11751         for generic 4->6.
11752
11753 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11754
11755         PR target/81616
11756         Generic tuning.
11757         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
11758         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
11759         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
11760         cond_taken_branch_cost 3->4.
11761
11762 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
11763
11764         PR tree-optimization/83581
11765         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
11766         TODO_cleanup_cfg if any changes have been made.
11767
11768         PR middle-end/83608
11769         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
11770         convert_modes if target mode has the right side, but different mode
11771         class.
11772
11773         PR middle-end/83609
11774         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
11775         last argument when extracting from CONCAT.  If either from_real or
11776         from_imag is NULL, use expansion through memory.  If result is not
11777         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
11778         the parts directly to inner mode, if even that fails, use expansion
11779         through memory.
11780
11781         PR middle-end/83623
11782         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
11783         check for bswap in mode rather than HImode and use that in expand_unop
11784         too.
11785 \f
11786 Copyright (C) 2018 Free Software Foundation, Inc.
11787
11788 Copying and distribution of this file, with or without modification,
11789 are permitted in any medium without royalty provided the copyright
11790 notice and this notice are preserved.